Note : Les descriptions sont présentées dans la langue officielle dans laquelle elles ont été soumises.
~O~ Z~ 4
This is a division of Canadian patent application
Serial No. 281,336 which was filed on 24 June 1977.
Field of the Invention
This invention relates to audio response units, and
more particularly, to memory arrays utilized in audio
response units for the storage of digitized speech.
Description of the Prior Art
The expanding use of computers has resulted in an
acute need to simplify and improve the communications link
between man and machine. Progress towards achieving
better communications between man and machine has been
made in the area of audio response units. These units,
under the control of an associated computer, store human
speech, either in analog or digital form, and respond to
information requests to the computer by compiling an audio
response which is forwarded to the human inquirer.
Responding in this manner to a human request eliminates
the need for teletypewriter terminals, CRT screens and
similar terminal apparatus and presents the requested
information to the user in a form with which the human
user is most familiar.
Widespread use of audio response units is contingent
upon the storage of a large vocabulary of words or
phrases. This was accomplished in the prior art by
storing individual words or phrases in analog form on a
magnetic drum and then selecting certain of these words or
phrases to form complete sentences. It was discovered
that a more efficient means of storing speech was to
digitize it before storage. Towards this end, prior art
systems have been developed in which analog speech is
. __
lO~Z~
digitized by applying the speech to an analog-to-digital
converter by dividing the digitized speech into segments
called grunts, which are generally shorter than most
words, and by storing the grunts on magnetic drums,
discs or tapes. The grunts are then retrieved from
storage in a predetermined sequence and applied to a
digital-to-analog converter to form analog words and
sentences.
The use of magnetic drums, discs or tape to
store digitized speech presents a number of difficulties
due to their weight, size and maintenance limitations.
In an attempt to eliminate these difficulties, it has
been found that solid state storage is an advantageous
medium for storing digitized speech in that it is of
minimum size and weight, has no moving parts and can
generally be readily adapted to even the most extreme
environments. Various types of solid state storage are
known in the art, such as Random Access Memories (RAM)
and Read Only Memories (ROM). Digital information stored
in a ROM is inserted therein during fabrication and is
not thereafter readily alterable. Digital information
stored in a RAM, however, is addressable, and can be
readily changed when necessary. Therefore, it is
advantageous to use a RAM in audio response units when the
stored vocabulary must be frequently changed.
The ability to frequently change the vocabulary
of an audio response unit is important in many applications
and therefore the technique of loading new digital informa-
tion into a RAM is an essential element in increasing the
usefulness of an audio response unit. RAMs are generally
arranged as a matrix array with a predetermined number of
--2--
X
- 10!~4224
intersecting rows and columns. Each row and column
intersection comprises a storage cell which stores one
bit of digitized speech. Conventionally, the number of
cells in each row of a memory array is limited to an
amount which is fewer in number than the number of bits
in a grunt. It is thus advantageous to store grunts in
columns of the memory array.
Storing a grunt in a column of the memory array
cannot be accomplished directly as conventional RAMs
are addressable on a row-by-row basis. The prior art
solves this problem by interleaving the bits of a
plurality of the incoming grunts so that loading the
memory array one row at a time results in storing each
of the grunts in one of the columns of the memory array.
More particularly, the computer associated with the memory
divides a stream of digitized speech intoi~-grunts, and
interleaves the bits of the various grunts to form a
serial data stream. The serial data stream is divided
into portions, with each portion consisting of a number
of bits equal to the number of storage positions in one
row of the RAM. The first portion of the data stream is
comprised of all the first bits from a number of different
grunts equal in number to the number of columns in the
memory array. The second portion is comprised of all the
second bits from the same grunts, the third portion is
comprised of all the third bits, etc. Successive portions
are shifted in series into a serial to parallel converter
and then written in parallel into successive rows of the
RAM. Each bit in each portion is associated with a
particular column of the array such that loading the
portions in parallel into successive rows of the array
.~
` 10~4Z24
results in storing each grunt in a particular column of
the array.
The above type of loading technique presents
problems when a particular data stream portion being
clocked into the serial to parallel converter gains or
loses a clock transfer pulse. This results in shifting
the entire stream of multibit portions one position left
or right such that the bits of all subsequent portions
clocked into the serial to parallel converter will also
be shifted. Thus, the bits of all portions subsequent
to the error will be written into a memory column which
is one position left or right from the column in which
the bit belongs. Therefore the occurrence of one clock
error during loading will result in each bit in all grunts
subsequent to the error being stored in an incorrect
column in the memory. This problem, known as "memory
streaking" in the prior art, requires a reload of the
entire memory each time it occurs. From the foregoing,
it is clear that the prior art method of loading results in
catastrophic failure upon the happenstance of a simple
clock error. This susceptibility to failure is a direct
result of the fact that the prior art method of loading must
interleave the bits of the incoming grunts in order to
load the grunts in the columns of the memory array.
It is therefore an ob~ect of this invention to
improve the technique of loading digitized speech into a
memory array.
Once digitized speech has been loaded into a
RAM without the occurrence of errors, it is essential
that a maintenance scheme be employed to periodically
check the sanity of the memory. Prior art maintenance
- 10!~4224
schemes normally consist of the associated computer simply
doing a parity check on each row of the RAM and indicating
a memory malfunction each time a parity error is found.
This procedure although accurate requires a substantial
amount of computer time.
It is therefore another object of the invention
to minimize the computer time required to determine
memory sanity.
An additional area of concern in the utilization
of a solid state memory to store digitized speech is the
reading out of the stored grunts from the memory to form
audio speech messages. In the prior art each row in the
array is read out and the bits from each row are distributed
to time slots in frames on a high speed time-division highway.
The stream of data on the time-division highway is routed
past a number of port circuits. Each po~t circuit is
provided with time slot addresses by the associated computer
which define ti~e slot positions in the data stream contain-
ing the bits of the grunts to be read f~om the memory array.
Each port circuit compares the time slot addresses with
time slot positions and each time a comparison is detected,
the port circuit extracts a bit from the highway and
applies the extracted bit to a digital-to-analog converter
thereby converting the binary grunts to analog speech.
A plurality of port circuits can be connected to the time-
division highway with each port circuit providing a
separate audio output. This prior art scheme requires
complex comparison circuits thereby rendering this scheme
expensive and complicated.
It is therefore another object of the invention
to simplify the manner in which digitized speech is read
out from a memory array.
--5--
10~4Z24
In accordance with one aspect of the invention there
is provided in a memory array comprising a plurality of
intersecting rows and columns, each row and column
including a plurality of binary storage cells, each
storage cell storing one binary bit, means for checking
the parity of the bits stored in each row of the memory
array and means responsive to the checking means for
indicating a malfunction in the memory array,
characterized by means responsive to the checking means
for counting the number of rows in which parity errors
occur, and means responsive to the counting means for
normally disabling the indicating means and for enabling
the indicating means at the time the number of rows
containing parity errors exceed a predetermined threshold
number.
In accordance with another aspect of the invention
there is provided a method for detecting a malfunction in
a memory array, the memory array consisting of a plurality
of intersecting rows and columns, each row and column
including a plurality of binary storage cells, each
storage cell storing one binary bit, the method comprising
the steps of, checking the parity of the bits stored in
each row of the memory array, counting the number of rows
in which parity errors occur, and indicating a malfunction
at the time the number of rows containing parity errors
exceed a predetermined threshold number.
- 5a -
109~224
In accordance with one aspect of the invention,
incoming digitized grunts are loaded into columns of the
memory array one at a time rather than several grunts
simultaneously.
It is a feature of the invention that the incoming
grunts in the form of a sequence of bits are loaded into the
memory array in a plurality of repetitive loading sequences,
each loading sequence being associated with an individual
column of the memory array and including reading out the bits
stored in sucessive rows of the memory array, replacing in
each successive row of bits, the bit from the associated column
with a successive one of the bits in the string of bits and
writing the read out row of bits, including the replaced bit,
back into the row. Loading grunts one at a time into columns
of the memory array solves the memory streaking problem
inherent in the prior art, since a bit shift in the incoming
stream due to the gain or loss of a clock transfer pulse simply
results in the bits subsequent to the error being shifted up or
down one position in the column. A shift up or down one
position in the column results in one incorrect bit in the
grunt stored in the column which is undetectable when the grunt
is reconverted to audio speech. This is in contrast to the
prior art in which a bit shift in the incoming data stream
results in bit shifts to other columns and thus in cata-
strophic loading failure.
It is another feature of the invention that each read
out row of bits is stored in a recirculating register, and the
bit read out from the associated column
1094Z24
is precluded from recirculating. The precluded bit is
replaced and the read out row of bits and the replaced bit
are written back into the row, whereby all bits except the
precluded bit are written unchanged back into the row.
It is a further feature of the invention that, since a
certain number of errors are acceptable in digitized
speech without degrading the audio output, memory sanity
is determined in accordance with the invention by counting
the number of rows in the memory array in which parity
errors occur and by indicating a memory malfunction when
the number of rows containing parity errors exceed a
predetermined threshold number. Indicating memory
malfunction only when parity errors exceed a predetermined
theshold minimizes the computer time necessary to
determine memory sanity.
In accordance with another aspect of the invention the
bits of grunts to be extracted from the memory array are
distributed to time slots in frames on a time-division
highway, each frame being of n bits duration and including
one bit from each of n grunts stored in the memory array.
It is another feature of the invention that a selected
grunt is extracted from the highway by commencing a count
of the bits on the time-division highway with a first bit
of the selected grunt and by extracting the first bit and
every nth bit thereafter.
The present invention taken in conjunction with the
invention disclosed in Canadian patent application Serial
No. 281,336 which was filed on 24 June 1977 will be
discussed in detail hereinbelow with the aid of the
accompanying drawings, in which:
10942Z4
Brief Description of the Drawings
In the drawings:
FIG. l discloses in block diagram form an
inquiry/response communications system containing an
audio response unit;
FIG. 2 discloses in block diagram form the
components of the audio response unit;
FIGS. 3A and 3B disclose a funcitional representa-
tion of a memory array utilized in the audio response unit;
FIG. 4 discloses the details of a port circuit
utilized in the audio response unit;
FIG. 5 discloses a timing diagram useful in
understanding the manner in which information is extracted
from the memory array;
FIG. 6 discloses a timing diagram useful in
understanding the manner in which information is loaded into
the memory array;
FIGS. 7 and 7A disclose a timing diagram for a
port circuit;
FIG. 8 discloses the details of a parity logic
circuit utilized in the audio response unit;
FIG. 9 discloses a timing digaram for the
parity logic; and
FIG. lO discloses the details of control
logic 215 shown in block diagram form in FIG. 2.
Detailed Description
Refer to FIG. l. Therein is shown a block
diagram representation of an inquiry/response communications
system designed to provide communications between a
plurality of remote terminals and a plurality of data
base computers for the purposes of credit verification,
X
10~42?~
point of sale transactions, electronic funds transfer, or
other inquiry/response applications. Data base computers
101 through 103 could, for example, be general purpose
computers located at a bank and would function to store
credit information or other information of interest to
a plurality of users. Remote terminals 118 through 120
could, for example, be standard PUSH BUTTON telephones or
alternatively could be inquiry/response terminals known in
the art. These remote terminals would be located on a
merchant's premises or other location from which it is desired
to access credit information or other information stored in
the data base computers. Central station 100 functions to
selectively connect any of the data base computers to any
of the remote terminals in a manner to be detailed herein-
after and contains a general purpose computer and associated
data communications equipment. Audio response unit 109 stores
digitized speech messages in accordance with this invention
and functions to transmit audio response messages to the
remote terminals under the command of the general purpose
computer located in the central station, as will be here-
inafter decribed.
A remote terminal desiring to communicate with
a data base computer places a call via telephone lines 115,
116, or 117 and telephone central office 114 to central
station 100. Subsequent to the completion of the call,
a remote terminal transmit signaling information via lines
115, 116 or 117, central office 114, and signaling paths
110 and 111 to the general purpose computer
located in the central station. The signaling
_g_
~. .
10~4ZZ4
information is used to identify which data base
computer the remote terminal wishes to communicate with.
Central station 100, in response to the signaling
information, connects signaling paths 110 and 111 to
two-way signaling paths 104, 105 or 106 and completes
a connection between the remote terminal and the designated
data base computer. Subsequent thereto, the remote
terminal transmits inquiries to the data base computer.
In response thereto, the data base computer processes the
inquiry and formulates a response, the response for
example, being the credit status of a particular customer
desiring to purchase an item at the merchant's premises
where the remote terminal is located. This response is
sent in digital format to central station 100 via signaling
paths 104, 105 or 106, and informs the central station
which type of audio reply should be returned to the
remote terminal. The central station transmits commands
to audio response unit 109 via buses 107 and 108 and
instru~ts the audio response unit to retrieve a particular
message from storage and transmit the message via paths
112 and 113 and central office 114 to the remote terminal
which initiated the communication sequence.
Audio response unit 109 stores a plurality of
audio response messages for transmission to the remote
terminals. These messages are made up of sentences or
phrases which are in turn made up of individual spoken
words. Each word is comprised of speech segments called
"grunts" which in general are shorter than a spoken word.
These elementary speech segments are digitized and stored
in a Random Access Memory (RAM) included in audio response
unit 109 for subsequent retrieval as will be described
in detail hereinafter.
..~; .
10!~4Z24
Refer to FIGS. 3A and 3B. FIG. 3A is a block
diagram representation of the RAM used in audio response
unit 109. The RAM is divided into 16 individual memory
sections. Each memory section in turn is comprised of
13 memory planes and each memory plane contains 32,768 bits
of binary storage. Each memory plane stores 8 grunts
with each grunt occupying 4,096 bits of storage.
.,
~09-~ZZ4
FIG. 3B shows a graphical representation of one
of the memory sections in the RAM. Each memory section
is organized in the same manner as is illustrated in
FIG. 3B. The memory section shown in FIG. 3B is
organized into a number of intersecting rows and columns.
More particularly, each memory section contains 13
columns of storage labled as memory planes 0 through 12,
and each memory section contains 32,768 rows of storage
labeled as addresses 0 through 32,767. Each memory
plane therefore contains 32,768 storage positions and
each row contains 13 storage positions with each
storage position containing one bit of digitized speech.
Each plane in the memory section contains 8 grunts with
each grunt occupying 4,096 bits of storage. Plane 0,
for example, contains grunts 0 through 7, plane 1 contains
grunts 8 through 15, etc. with plane 11 containing
grunts 88 through 95 and plane 12 containing grunts 96
through 103. Grunts are stored in each plane in
asc~nding bit order. For example, positions 0 through 7
of Elane 0 contains bit 0 from the 8 grunts stored in
plane 0. Positions 8 through 15 of plane 0 contains
bit 1 from the 8 grunts stored in plane 0, etc. with
positions 32,760 through 32,767 of plane 0 containing
bit 4,095 of the 8 grunts stored in plane 0. The
remaining planes of the RAM are organized in an identical
manner as is shown in FIG. 3B.
The information stored in each section of the
RAM can be accessed on a row-by-row basis. For example,
when address 0 is applied to a memory section, all the
bits stored in row 0 are accessed, that being bit 0 of
grunt 0, 8, 16, etc. up to grunt 96. Address 1 accesses
-12-
i.O~ 2Z4~
bit 0 of grunts 1, 9, 17, etc. up to grunt 97. The remain-
ing rows of each memory section are accessed as is shown
in FIG. 11.
FIGS. 3 and 11 illustrate a functional represent-
ation of a RAM emphasizing the matrix array arrangement.
It is to be understood that such a RAM would require input
and output circuitry, refresh circuitry etc. to allow information to be
entered into and extracted from the RAM. Such circuitry is
well known in the art. In all subsequent references to the
RAM, it is assumed that such circuitry is included in the
block diagram representations of the RAM and functions to
enter information into and extract information from the RAM
in a manner which is well known in the art.
Refer to FIG. 2. Therein is shown the details of
audio response unit 109. Incoming data bus 107 from the
processor in central station 100 is appliëd to input terminal
201. This data bus carries information to be stored in the
RAM and carries grunt number identification information
as will be detailed hereinafter. Incoming mode control
bus 108 from the processor in central station 100 is -
applied to input terminal 200 and carries command
information for the audio response unit as will be described
hereinafter. This command information is applied to
logic 202 and logic 204. Logic 202 in response to the
command information from central station 100 generates
address and command information for the memory sections
of the RAM and generates command information
'X
~O!~Z2~
and a clock signal for the memory input/output logic~
Address and command information for the RAM sections
are applied to the sections via buses 207 and 208
respectively. This information is of the type necessary
to address specific rows of storage in the RAM and to
command the RAM sections to accept information from the
input/output circuitry for storage in a particular row
or to present information from a specific row to the
input/output circuitry. The logic necessary to generate
this address and command information in response to
command information from the processor is dependent on
the specific type of RAM used in the audio response
unit. Given aspecific RAM the design of this logic
would be obvious to one skilled in the art. This logic
forms no part of the invention described herein and
therefore that portion of logic 202 which generates
address and command information for the RAM sections
in response to command information from the processor
will not be described in further detail. Logic 202 also
generates a clock signal which is applied to the input/
output circuitry and logic 215 via`path 206. The clock
necessary to generate this signal is a free running
oscillator of the type well known in the art and will
not be described further. Logic 202 also generates a
load port circuit control signal which is applied to
logic 215 via path 218. The remaining signal generated
by logic 202 is a load command signal applied to the
input/output circuitry via bus 205. The relationship
of the load command signal and the load port circuit
control signal to the other signals utilized in the audio
response unit is shown in FIG. 5 and will be discussed
-14-
. ~ .
lO9'~ZZ9~
hereinafter. The logic necessary to generate these signals
will be obvious to one skilled in the art when the relation-
ship of these signals to the other signals is discussed in
conjunction with FIG. 5. Load logic 204 accepts mode control
commands from central station 100 and in response thereto
generates two control signals, the enable load signal, and the
enable recirculate signal. These control signals are applied
to each memory input/output circuit via signaling paths 230
and 231'and will be described in detail hereinafter. Each
memory section 0 through 15 has associated therewith memory
input/output circuitry shown as memory input/output 210
associated with memory section 0, through memory input/output
226, associated with memory section 15. Each of the remaining
memory sections 1 through 1'4 have associated therewith an
identical memory input/output circuit. ~lemory input/output
circuit 210 and the associated input/output circuits function
to extract information from the memory sections and function
to store information in each memory section. The input/output
circuits are controlled by commands from memory address
counter and control logic 202, and load logic 204 in a manner
to be detailed hereinafter. Grunt'buffer and control logic
215 accepts grunt number addresses (i.e., the address of a
particular grunt stored in the memory sections) from central
station 100 and applies these addresses to bus 216. Control
logic 215 also generates certain clock and control signals
and applies these signals to bus 217. The details of control
logic 215 will be described hereinafter. Bus 216 is extended
to port circuits 0 through 111, and bus 217 is extended to
-15-
Z24
port circuits 0 through 111 and to parity logic 221. Each
port circuit 251 accepts information from each memory input/
output circuit, and converts this information into audio
response messages as will be detailed hereinafter. The
audio response messages are applied to output leads 222, 223,
etc. Parity logic 221 functions to perform two types of
parity checks on the contents stored in each memory section.
The manner in which these parity checks are accomplished will
be detailed hereinafter.
Refer to FIG. 4. Therein is shown the details of
one port circuit such as port circuit 0. The remaining port
circuits are arranged in an identical manner. Data buses
234 through 250 extending from each memory input/output
circuit are applied to input terminals 400 through 415 and
extended to selector 418. Grunt number bus 216 is applied
to input terminals 425 through 428 and to input terminals
433 through 439. The 4 bits appearing on input terminals
425 through 428 are stored in 4-bit latch 419 and are
applied to selector 418. In response thereto, selector 418
selects one of the 16 incoming data buses defined by the
4 bits and applies that selected data bus to lead 440 extend-
ing to flip-flop 422. The bits appearing on input leads 433
through 439 are applied to programmable counter 421 and
function to presetcounter 421 to a particular state out of
112 states defined by the incoming bits. Also applied to
counter 421 is a port clock, load and enable signals
extending from port control bus 217 via input terminals 429
through 432 and input terminal 417. The function of
counter 112 will be described herei-after. D/A
converter 423 and its associated circuitry functions
-16-
~O~Z24
to convert the digitized speech extracted from the RAM
into an audio output which is appliecl to lead 222 and
transmitted to the remote terminals. The function
performed by the D/A converter and its associated
circuitry of converting a serial digital signal into an
audio signal is well known in the art and therefore
will not be described in further detail.
Refer to FIGS. 2 and 5. The method of extracting
information from the R~M will now be described in detail.
Waveform A in FIG. 5 is the shift clock pulse train
generated by logic 202 and applied via path 206 to the
clock input of register 212 and to the remaining input/
output circuits. The shift clock waveform is a free-
running clock continuously applied to the input/output
registers. Extracting information from a memory section
begins with a load command from logic 202 applied to
gates 211 and register 212 in each memory section via
lead 205. This load command is shown as Waveform B in
FIG. 5 and is generated by logic 202 in response to a
command applied to logic 202 from the processor in
central station 100 via bus 108 and terminal 200.
Assume also that in response to the command from the
central station processor, logic 202 at this same time
addresses row 0 in each memory section via buses 207
and 208 such that the information in row 0 of each memory
section will be presented to gates 211 in the memory
input/output circuitry. (It is to be understood that
logic 202 applies the address commands to every memory
section but as all memory sections operate in the same
manner, only memory section 0 will be described in
detail.) The load command applied to gates 211 enables
10~2Z4
these gates to transfer the information from row 0 in
the memory section to register 212 for storage. The
load command appleid to the LO/SHIFT input of register
212 places the register in the load mode to accept the
information from gates 211 and thereafter register 212
reverts to the shift made to allow the information
stored in the register to be shifted out. The
information stored in register 212 is clocked from this
register in response to the shift clock pulse train and
appears on data bus 234. Gate 214 is at this time
disabled as will be described hereinafter such that the
information is applied to the port circuits via bus 234.
Waveform D in FIG. 5 illustrates the format of the
information appearing on bus 234 as it is clocked out
of register 212. As is shown therein, the information
consists of the bits from planes 0 through` 12 of the
memory section with the bits being bit 0 of grunt 0,
bit 0 of grunt 8, etc. Subsequent to loading row 0 into
register 212, an increment memory pulse (Waveform C in
FIG. 5) is applied via lead 207 to memory sections
0 through 15. In response thereto, each memory section
applies the contents of row 1 from the section to the
memory input/output circuit subsequent to the increment
memory address pulse, an additional load command from
control logic 202 is applied to gates 211 and register 212
in each memory input/output circuit. In response thereto,
row 1 from each memory section is then stored in
register 212.
-18-
X
~094Z~
This process of incrementing the memory address and extracting
a particular row from a memory section continues until all
the information stored in each memory section has been
applied to the output data bus of each memory input/out
logic circuit. The format of the data on a particular data
bus can be illustrated by referring to Waveforms E and F in
FIG. 5. As is shown therein, segment 0 of waveform E
contains the 13 bits of information from row 0 of a
particular memory section. For example, segment 0 contains
bit 0 from grunt 0, 8, 16, etc. up to grunt 96. The 14th
bit (labled S) shown in segment 0 is reserved for control
purposes. Section 1 of Waveform E contains the contents of
row 1 from a memory section. For example, segment 1 contains
bit 0 from grunts 1, 9, 17, etc. up to grunt 97. The
remaining segments of Waveform E contain bit 0 from the
remaining grunts stored in the memory section. As is shown
in FIG. 5, Waveform E illustrates the 112 bits resulting
from extracting bit 0 from all 104 grunts stored in a
memory secti(n and applying these bits to the memory input/
output data buses. Therefore, Waveform E contains 1 bit
(bit 0) from each grunt stored in a memory section. This
segment of 112 bits will hereinafter be referred to as a frame
of data. Waveform F in FIG. 5 is a representation of the
data bus format containing all bits from all grunts stored
in a particular memory section. This waveform is therefore,
comprised of 4,096 frames of data with each frame containing
1 bit from each grunt stored in a memory section. For
example, frame 0, of Waveform F corresponds to Waveform E
and shows the position of bit 0 from each grunt on the data
bus. Frame 1 of Waveform F shows the position of bit 1,
frame 2 shows the position of bit 2, etc., up to frame 4,095
. .
~ ~ ,.
10~?4Z24
which shows the position of bit 4,095 of each grunt on the
data bus. As is shown in FIG. 5, Waveform F illustrates
the data format when each bit of all 104 grunts stored in a
memory section are applied to the input/output circuit data
buses. The combination of all 4,096 frames of data on the
data bus will hereinafter be referred to as a superframe
of data.
The foregoing has illustrated how information
stored in each memory section has been applied to the data
bus and in turn to the port circuits. It is now necessary
to extract certain bits from that information in order to
form response messages to the remote terminals. Extracting
this information is accomplished by port circuits 0 through
111 in response to grunt addresses applied to the port
circuits from grunt buffer and control logic 215.
Refer to FIG. 5. Waveform G in FIG. 5 is the
load port circuit control waveform generated by
logic 202 and applied to logic 215 via path 218. This
waveform is normally high and goes low simulta~eously with
frame 4,095 illustrated in Waveform F. Therefore, this
waveform is high throughout the entire superframe of data
on the data bus and goes low for the last frame of the
superframe. Waveform G is extended as is shown in FIG. 5 to
the time interval illustrated by Waveform H. Waveform I in
FIG. 5 occupies this same time interval. Waveform I
illustrates the grunt address information applied to grunt
number bus 216 during the interval illustrated by Waveform
H. As illustrated in Waveform I the grunt address informa-
tion is serially applied to the grunt bus by logic 215 with
the grunt number for port 0 appearing first, the grunt
number for port 1 appearing second, etc. This information
-20-
109~Z?~4
is utilized to command the port circuits to extract certain
grunts from the data buses as will be illustrated hereinafter.
Refer now to FIGS. 4 and 7. From what precedes,
it has been shown that grunt address information is applied
to grunt bus 216 and to each port circuit during the last
frame of the superframe existing on the data buses. The
grunt address information is stored in the grunt buffer by
the processor in central station 100 and applied to the grunt
bus in a manner to be detailed hereinafter. This grunt
address information is applied to input terminals 425 through
428 and input terminals 433 through 439 in FIG. 4. This
information will now be used to select one of data buses 0
through 15 and to select a particular grunt from the super-
frame of data on the data buses.
Waveforms B through M in FIG. 7 illustrates waveforms
generated by control logic 215 and appliedr-to the port
circuits via bus 217 in a manner to be detailed hereinafter.
Waveform N in FIG. 7 illustrates the grunt addresses applied
~o bus 216 (FIG. 2) by logic 215. Each interval PO through
Plll illustrates the interval each grunt address occupies
on bus 216 wherein PO is the grunt address for port circuit
0 and Pl is the grunt address for port circuit 1, etc.
Waveform A in FIG. 7 illustrates the load port circuit
control waveform generated by logic 202 and designated as
Waveform H in FIG. 5. More particularly, Waveform A illustrate
the interval corresponding to the interval occupied by
frame 4,095 in the superframe of data on the data buses.
Waveforms B and K in FIG. 7 are port control commands
applied to input terminals 429 on selected ones of the port
circuits (FIG. 4). Waveform B in FIG. 7 is applied to port
circuits 0 through 15. Waveform K is applied to port
ZZ4
circuits 16 through 31. Not shown in FIG. 7 are five
additional waveforms which perform the same function as
waveform B and K. These waveforms are applied to input
terminal 429 of port circuits 32-47, 48-63, etc. Waveforms
C through J are applied to input terminals 430 on selected
ones of the port circuits. Waveform C in FIG. 7 is
applied to port circuits 0 and 1, 16 and 17, etc. Waveform
D is applied to port circuits 2 and 3, 18 and 19, etc.,
and Waveform E is applied to port circuits 4 and 5, 20 and
21, etc. Waveforms F through J are applied to the
remaining port circuits in the same manner as discussed for
Waveforms C through E. Waveforms L and M in FIG. 7 are
applied to input terminals 431 on alternate port circuits
as is indicated in FIG. 7. Refer now to FIG. 4 and assume
that this figure represents port circuit 0. As has been
described above, in accordance with FIG. 1, the grunt
addresses on grunt bus 216 are applied to each port circuit
during frame 4,095 of the superframe. From FIG. 7 it can be
seen that Waveforms B, C and L are ail low during the
interval occupied by grunt address PO on bus 216. As
described above, Waveforms B, C and L are applied to input
terminals 429 through 431 on port circuit 0. In response
thereto, gate 420 (FIG. 4) is enabled which in turn enables
latch 419. Latch 419 applies 4 bits of information from the
grunt bus to selector 418. As described above these 4 bits
of information define a particular one of the data buses
applied to input terminals 400 through 416 on the port
circuit. In response thereto selector 418 selects one of
the 16 data buses defined by the last 4 bits of the grunt
address and applies that data bus to path 440 and in turn to
flip-flop 422. The output of gate 420 also applies a load
-22-
X
4ZZ4
command to the load input of counter 421. (Counter 421 is
disabled by Waveform A FIG. 7 being applied to the EN input
and therefore is not counting at this time.) The load
command applies the remaining 7 bits of the grunt address
to the inputs of counter 421. These 7 bits preset counter
421 to a particular state defined by the 7 bits from among
the possible 112 states in counter 421. Port circuit 1
would be similarly enabled by Waveforms B, C and M such that
port circuit 1 would select one of the 16 data buses, and
the counter in port circuit 1 would be preset to a state
defined by grunt address Pl on bus 216. The remaining
port circuits would be preset in the same manner. At the
conclusion of frame 4,095 in the superframe therefore, each
port circuit has been present to select one of the 16 data
buses and preset to select one of a possible 112 states.
Referring now to FIG. 5, Waveform E, it can be seen that
each frame of the superframe contains 112 bits with each
bit (except the control bits) being one bit from a particular
grunt stored in the ~A~i,
Refer now to FIGS. 4 and 7A. Waveform A in FIG. 7A
illustrates the load port circuit control waveform generated
by logic 202 which has been previously described in FIGS.
5 and 7. This waveform is applied to input terminal 432
of each port circuit and when this waveform goes high, it
enables counter 421.~ Counter 421 in response to the port
clock appearing on terminal 417 will begin counting at this
time at the state that has been preset by the first 7 bits
of the grunt address. This counter will then count from
that preset state to state 112 at which time the Q output of
counter 421 will go high and be applied to the toggle input
of flip-flop 422. This flip-flop will then extract one bit
,~,
Jb0~4~:24
of data appearing on lead 440 and apply that bit to
D/A converter 423. Referring to Waveforms B and C in
FIG. 7A, it can be seen that Waveform B represents the
superframe on each of the data buses. This superframe
of data is identical to the superframe shown in
Waveform F in FIG. 5. Waveform C in FIG. 7A illustrates
that counter 421 begins counting when enabled by the load
port circuit control pulse and counts 112 minus the grunt
number clock cycles at which time one bit of data from
the superframe is strobed into flip-flop 422 and applied
to D/A converter 423. Counter 421 then resumes counting
for an additional 112 clock cycles at which time a second
bit of data will be strobed from the superframe and
applied to flip-flop 422. This process continues with
counter 421 strobing every 112th bit from the superframe.
Referring now to Waveforms E and F (FIG. O , it can be
seen that counter 421 if preset to begin with bit 0 of
grunt 0 will count 112 clock cycles and then strobe bit 1
of grunt 0, count an additional 112 cycles and strobe
20 bit of grunt 0, etc. Similarly counter 421 could be
preset to begin with bit 0 of any grunt stored in the
RAM and would thereafter extract all the bits of the
grunt. In this manner, therefore, presetting each
counter in each port circuit allows each port circuit
to extract from the data buses the bits from any partic-
ular grunt stored in the RAM and apply the bits from
that grunt to the D/A converter and its associated
circuitry which will in turn reconvert the bits of the
digitized grunt into an audio response message.
Therefore extracting information from the random access
memory is simply accomplished by presetting each port
-24-
lO~ZZ~
circuit with a particular grunt address and then
allowing that port circuit to re~etitively extract from
the data buses each bit from that particular grunt and
convert that grunt into an audio response message.
The preceding has described the procedure for
extracting digitized speech from the RAM and converting
that digitized speech into audio response messages. The
procedure for loading digitized speech into the RAM
will now be described in detail. Various procedures
exist for converting audio speech into digitized speech
for insertion into a solid state memory. Such conversion
procedures are not part of this invention. Therefore
it will be assumed that apparatus which is not part of
this invention is included in central station 100 and
accepts audio messages to be stored in the audio
response unit.
-25-
~O~ ~Z29t
The apparatus is advantageously designed to convert the
audio messages into a digital pulse train ready for
storage in the RAM. This digital pulse train would then
be applied to the general purpose computer in central
station 100. The general purpose computer is programmed
to divide the digital pulse train into serial grunts. The
serial grunts are then applied via data bus 107 to input
terminal 201 of the audio response unit. The general
purpose computer in central station 100 is programmed to
distribute the various grunts to the memory sections in
which the grunts are to be stored. More particularly
bus 107 extending from the computer is 16 leads wide with
1 lead going to gate 213 in each memory input/output
circuit via bus 232. The computer assembles the grunts
in serial form and applies the serial grunts simultaneously
to the correct memory input/output circuit via the separate
leads of bus 107 and bus 232. At the time central station
100 applies the grunts to data bus 107, central station 100
also applies command information to mode control bus 108.
The command information would be applied to terminal 200
and from there to control logic 202 and load logic 204.
The command information can be of any suitable format
and functions to instruct control logic 202 and load
logic 204 that incoming grunts have been applied to data
bus 107.
Refer to FIG. 6. The command information to
instruct control logic 202 that serial grunts have been
applied to data bus 107 is assumed to take the form of
Waveform D in FIG. 6. The load data pulse in Waveform D
is applied to input terminal 200 and from there to control
logic 202. In response thereto, control logic 202 will -
-26-
X
.
lO~Z~4
supply address and command information to each memory
section, instructing each memory section to read out
the data stored in row 0 and apply that data to the
inputs of gates 211. Control logic 202 will also generate
Waveform B in FIG. 6 and apply this waveform via path 205
to gates 211 in memory input/output circuit 210 and of
course to the remaining input/output circuits. In response
thereto, gates 211 in each memory input/output circuit
will gate the first row of information stored in the
memory sections from the memory into register 212 in each
memory input/output circuit. The information stored in
register 212, from memory section 0, is represented by
Waveform C in FIG. 6, and will be shifted from register 212
by the shift clock and will be applied to path 234 and 219
to the input of gate 214. The same process occurs for
each memory input/output circuit. The information shown in
Waveform C as segment 0 represents the information stored
in row 0 plane 0 of memory section 0, segment 1 represents
the information stored in row 0 plane 1, etc. In this mann~r,
the information stored in register 212 for each memory
input/output circuit is recirculated from the output
of register 212 back to the input of register 212.
Recall that central station 100 is attempting to
store incoming grunts in each memory section. The grunts
are in serial form and are to be stored in each memory
section beginning with plane 0 and progressing through
plane 12. More particularly, referring to FIG. 3B, it
can be seen that the stream of serial grunts would begin
with bit 0 of grunt 0 which would be stored in plane 0
row 0, followed by bit 1 of grunt 0 which would be stored
in plane 0 row 8, etc. The bits of grunt 0 would be
followed by the bits of grunt 1, etc. The serial stream
-27-
'
~o~
of grunts for each memory section is organized in the samemanner and each stream consists of the grunts to be
stored in a particular memory section.
Load logic 204 in response to the command
information applied thereto via bus 108 generates
Waveforms E and F (FIG. 6). These waveforms are applied
to gates 213 and 214 in each memory section via buses
230 and 231. As described above, the incoming grunts are
to be stored in each memory section beginning with plane 0.
As is shown in FIG. 6, Waveform E goes low and Waveform F
goes high in conjunction with the data in row 0 plane 0
appearing on the data bus. This data is identified
as segment 0 in Waveform C. At the time, segment 0 of
the information stored in register 212 appears on the
data bus it is applied to the input of gate 214 in each
memory section. The EN REC waveform going-low disables
gate 214 and the EN LOAD waveform going high, enables
gate 213. Therefore the information from row 0 plane 0
for each memory section is prevented from being clocked
into register 212 while at the same time, the first bit
of the incoming grunt appearing on bus 232 is applied to
the input of enabled gate 213 in each memory section.
This bit is clocked into register 212 in the position
previously occupied by the information that was stored
in row 0 plane 0 of each memory section. Subsequent to
segment 0 of Waveform C, the EN REC waveform returns high
and the EN LOAD waveform returns low. Therefore the
remaining information appearing on bus 234 is simply
recirculated from the output of the register back to the
input of the register via enabled gate 214. Subsequent to
segment 12 of Waveform C, all 13 bits originally stored in
-28-
' ~~
.... ~ .
10~34~?~
register 212 have been recirculated back to their
original positions. At this time, Waveform G goes high.
Waveform G is a load command from control logic 202
which is applied to gates 211 via path 205. In response
thereto, the information stored in register 212 is gated
from register 212 back into row 0 of memory section 0.
This information, as previously described, now contains
the first bit of the incoming grunt which was inserted
in register 212 in the position corresponding to plane 0
of row 0 in memory section 0. Therefore, when this
information is reloaded into the memory, the first ~it
of the incoming grunt is loaded into row 0 plane 0, while
the remaining bits previously stored in row 0 which have
not been changed, are simply reloaded unchanged into
their previous locations.
Subsequent to reloading the information into
the memory, Waveform H goes high. Waveform H generated
by control logic 202 is applied via address bus 208
to memory section 0 and to the other memory sections
and adds 8 to memory address. Therefore, row 8
of each memory section will be accessed next. Referring
to FIG. 3B, it can be seen that row 8 of memory section 0
contains the second bit of the grunts stored in the
memory section.
The procedure described above for loading the
first bit of the incoming grunt into the memory section
is now repeated with the information stored in row 8
of the memory section. This information from row 8 is
loaded into register 212, recirculated, and reloaded into
the memory after the information stored in row 8 plane 0,
has been replaced by the second bit of the incoming
-29-
~0!~4ZZ~
grunt. This process continues and functions to load the
incoming serial grunt, bit by bit, into plane 0 of each
memory section. This same process is repeated to load
each plane of each memory section.
In summary, a serial incoming grunt is applied
by central station 100 to the audio response unit via
bus 107. Simultaneous therewith central station 100
applies command information to logic 204 and logic 202 to
alert the logic that incoming grunts are to be stored
in the memory sections. Logic 202 and logic 204 supplies
address and command information to the memory sections
and their associated input/output circuit. In response
thereto the incoming grunts are loaded bit by bit into
the columns of the memory array. This process is
accomplished by reading out the rows of the memory
sections, recirculating the information s~ored in the
rows, replacing one of the bits of the recirculated
information with successive ones of the bits of the
inco~ing information, and reloading that information
into the positions previously occupied by this row of
information in the memory array. This process of loading
the information into the memory on a column by column
basis funetions to prevent the prior art problems of
memory streaking. Recall from the description of the
prior art that memory streaking occured in the prior art
when a word being loaded into a particular row of the
memory array was shifted one position left or right in
the input serial to parallel converter due to the gain
or loss of a eloek transfer pulse. This resulted in
shifting the entire ineoming stream of data one position
left or right such that all bits subsequent to the error
-30-
16)~4Z~
would be written into the wrong memory column. When
this occurred, it became necessary to reload the memory
due to the large number of errors contained therein.
In the instant invention a similar shifting of the
incoming data stream would occur if the computer in
central station 100 would transfer a bit to the audio
response unit via bus 107 and for one reason or
another would fail to command load logic 204 and control
logic 202 to load the bit into the memory array in the
manner described above. The result of such an error
with the instant invention, however, advantageously
does not result in catastrophic failure as it does in
the prior art. More specifically assume serial grunts
are being loaded into a particular memory section. The
format of the incoming data advantageously consists of
the bits of grunt 0 in sequence followed by the bits
of grunt 1 in sequence etc. If the computer failed to
provide the proper command when the first bit of the
first grunt was applied to bus 107 the first bit would
be lost. Assume now that the proper command was received
when the second bit was applied to bus 107. The second
bit would then be loaded into row 0 plane 0 of the memory
section which is the incorrect position for that bit.
Similarly, the third bit would be loaded into row 8
plane 0, the fourth bit would be loaded into row 10
plane 0 etc. All subsequent serial bits in the incoming
data stream would thus be shifted one position and would
be loaded into the wrong position in the particular
column into which they were being stored. The result
of the error therefore would be that the information
stored in each column of the memory array subsequent to
~'~'
~o~lz2~
the error would be shifted one position up or down from
its proper position in the columns. A shift up or down
one position in the columns results in each grunt being
shifted one position such that each grunt in a column
will have one incorrect bit.
As has been previously stated, one incorrect
bit in each grunt does not affect the audio output.
Therefore it can be seen that a shift in the incoming
data stream advantageously results in a limited number
of errors that are undetectable in the audio output.
In contrast, a shift in the incoming data stream in
the prior art causes each bit in each grunt stored
in the memory subsequent to the error to be incorrect
due to the row-by-row, loading method in the prior art.
~ .
~0~ ~2~
Refer to FIG. 8. Therein is shown the details
of parity logic 221 which is illustrated in block diagram
form in FIG. 2. Parity logic 221 functions to check the
sanity of the R~M by performing two types of parity
checks. The first type of parity check is designated as
a soft parity check. The soft parity check essentially
consists of checking the parity across each row of the
random access memory. Recall from what was described
above that a data bus from each memory section input/output
circuit is applied to the parity logic. Each data bus
carries a data stream as indicated in Waveforms E and F,
FIG. 5, wherein the information from each row of a memory
section is successively applied to the data bus. The soft
parity check procedure first selects four of the data
buses from among the sixteen available data buses. The
four data buses selected correspond to a~particular shelf
of the memory wherein a memory shelf consists of four
memory sections as is illustrated in FIG. 3A. A parity
check is performed on the 4 ~its which comprise row 0
plane 0, in the four memory sections. This process is
then repeated throughout the rows of the memory wherein
parity would be successively checked for row 0 plane 1,
row 0 plane 2, etc., until a parity check had been
completed on a particular row extending across one shelf
or four sections of the memory. The process continues
until parity is checked for each row in the entire memory
shelf.
The memory is arranged such that each row of a
particular shelf of the memory has even parity, i.e., an
even number of binary "l's" are stored in each row of a
memory shelf. (Note that each shelf consists of four
-33-
10S~4~24
sections so that one row from one memory shelf consists
of 52 bits.) If an odd number of binary "l's" are
detected in a particular row of a memory shelf it
indicates that one or more of the bits stored in that
row are incorrect. The soft parity technique counts the
n ~ ers of rows in each memory shelf in which odd parity
occurs. If the number of rows in which odd parity is
detected exceeds a predetermined threshold number, the
computer in central station 100 is notified of a memory
malfunction. The predetermined threshold number is set
by the computer and stored in the parity logic before a
parity check is performed in a manner to be detailed
hereinafter. Therefore, the computer will not be informed
of a memory malfunction until the number of rows in which
odd parity is detected exceeds the predetermined threshold
number which was previously set by the co~puter. Recall
from what was described above that digitized speech stored
in a random access memory can tolerate a certain number
of errors before the speech output is distorted. There-
2 n fore, with this method, the computer can determine thenumber of errors which can be tolerated in the memory
array and preset the parity logic to the threshold number
of errors. Once the parity logic is preset, the computer
will not thereafter be notified of a memory malfunction
until the preset threshold number is equaled or exceeded.
The second type of parity check performed by
the parity logic is a background type of maintenance and
is designated as a check sum parity check. This scheme
is utilized periodically, perhaps once a day, and is
normally accomplished when the memory is off line. This
procedure essentially consists of adding all the binary
-34-
~ .
2Z~
"l's" in each column of the memory array. The 8 least
significant bits of this sum for each memory plane is
then sent to the computer in central station 100. The
computer knows the number of binary "l's" that should
be stored in each plane. Sending the computer the 8
least significant bits defining the number of "l's" in
each plane allows the computer to note trends in the
sanity of the memory and could, for example, allow the
computer to detect gradual degradation in the memory
over a period of time. In addition to detecting trends
in memory sanity, this procedure is also useful when
utilized subsequent to the soft parity technique to
isolate a particular memory plane or planes that are
malfunctioning as will be detailed hereinafter.
The parity logic operation will now be described
in detail. Refer to FIG. 8. Input termi~al 800 carries
information from path 251 (FIG. 2). This information is
illustrated in FIG. 5, Waveform B, as the "load shift
register from memor~" waveform. This waveform is a~plied
to retiming circuit 801 which retimes Waveform B to
correspond to the logic operations performed by the parity
logic. Such retiming functions are well known in the art
and will not be discussed in further detail. The 16 data
buses from the memory sections are applied to selector 805
which functions to select four of the 16 buses in response
to address bits applied to selector 805 from selector 806.
Selector 806 selects either of input buses 822 or 823
and applies these buses to selector 806 as will be
detailed hereinafter. Shelf counter 824 functions to
select a particular memory shelf on which to perform
the soft parity check as will be subsequently described.
-35-
105`4Z%~
Latches 810 and 811 accept information from the computer
via bus 232 (FIG. 2). The information applied to latches
810 and 811 consists of either the predetermined number
of threshold parity errors which has been predetermined
by the computer of alternatively this information defines
a particular plane in the memory on which to perform the
check sum parity check. Input terminal 812 accepts
information from bus 217 in FIG. 2. This information is
the load port circuit control pulse illustrated as
10 Waveform G in FIG. 5. Input terminals 813 and 814 accept
information from the processor via bus 232 which determines
the mode of the parity logic, i.e., either the check
sum mode or the soft parity mode. Parity check circuit
817 operates during the soft parity mode and functions to
check the parity in each row of a memory shelf as will
be detailed hereinafter. Parity flip-floF-819 determines
odd or even parity for each memory shelf. Selector~818
functions to select a particular bus from among the 16
~ data buses and operates in con~unction with counter 828
to select a particular memory plane during the check sum
mode of operation. Counter 835 is an 8-bit counter in
which the predetermined threshold number of parity errors
is used to preset the counter to the threshold state.
This counter also functions to count the number of binary
"l's" in a particular memory plane during the check sum
mode of operation. Latches 838 and 839 store information
to be transferred to the computer via output drivers 840
and 846 as will subsequently be described. Mode circuit
816 accepts commands from the computer via buses 217 and
232 and generates selected output signals as will be
detailed hereinafter.
-36-
10'~'122~
Refer now to FIG. 9 and assume that the computer
in central station 100 desires to place the parity logic
in the soft parity mode. As described above, Waveform A
in FIG. 9 is the "load port circuit control" waveform
illustrated as Waveform G (FIG. 5). This waveform is
applied to terminal 812 of mode circuit 816. Mode
circuit 816 is enabled in response to this waveform
going low and generates certain output signals as will
subsequently be described. As circuit 816 is enabled
when this waveform goes low, it is to be noted that the
functions performed by mode circuit 816 occur during
frame 4,095 of the superframe in FIG. 5. Waveform B
in FIG. 9 represents a strobe command from the computer
in central station 100 applied to input terminal 807.
When Waveform B goes high, the computer loads 8 bits
into latches 810 and 811 defining the thr-ëshold number
of rows in the RAM in which parity errors can be detected
before a memory malfunction is indicated.
At the time the 8 bits are loaded into la ches
810 and 811 the computer applies Waveform C to input
terminal 813 thereby notifying circuit 816 to enter the
soft parity mode. In response thereto the CKSM output
of circuit 816 goes low and the IDLE output goes high.
The CKSM output going low applies a high to one input of
gate 833 via inverter 825. The remaining input of gate 833
also goes high in response to the IDLE output. Gate 833
then applies a LD (load) pulse to counter 835 which loads
the information stored in latches 810 and 811 into the
preset (PS) input of counter 835 via bus 841. These 8
bits preset counter 835 to a predetermined state equal in
number to 256 minus the threshold number of rows in which
-37-
~O!~lZZ4
parity errors can occur before memory malfunction is
indicated. The CKSM output of mode circuit 816 is also
applied to selector 806. In response thereto, selector
806 selects bus 823 extending from shelf counter 824.
Counter 824 which was initially cleared, applies 2 bits
via bus 823 to selector 806. These 2 bits which are
both logical "O's" (as counter 824 was cleared) are
applied to selector 806 and from selector 806 to selector
805. Selector 805, in response to these 2 bits, selects
data buses 0 through 3 extending from memory sections 0
through 3. Memory sections 0 through 3 comprise shelf 0
as is illustrated in FIG. 3A. Waveforms I through K in
FIG. 9 illustrate the data appearing on the four data
buses extending from memory sections 0 through 3. These
four data buses are applied to parity check circuit 817
which functions to determine even or odd parity across
each set of 4 bits appearing on the four data buses. As
is illustrated in FIG. 9, plane 0 of row 0 results in even
~parity. (This is indicated in FIG. 9 as an "E" appearing
directly beneath plane 0 of row 0.) Plane 1 of row 0,
however, results in odd parity. In response thereto,
parity flip-flop 819 is toggled to the SET state as is
illustrated in Waveform N. Plane 2, row 0 again results
in odd parity which toggles flip-flop 819 back to the
CLEAR state. This process continues until row 0 in all
four memory sections has been checked. If the bits in
row 0 from the four memory sections consist of an odd
number of binary "l's", flip-flop 819 will be in the SET
state at the conclusion of the parity check for that row.
At this time, Waveform M (the "load SR from memory" pulse
from buses 205 and 251, FIG. 2) is applied to input
-38-
2~ .
~0~4Z24~
terminal 800. This waveform goes high, which applies a
high to one input of gate 820. The remaining inputs of
gate 820 are also high as CRSM is high add flip-flop 819
is in the SET state. In response thereto, a pulse is
applied via gate 826 and enabled gate 834 (COUNT is high)
to the EN input of counter 835. Counter 835 is therefore
clocked to the count of "1", indicating that there was
a parity error in the first row of the memory shelf.
This process continues for each row of the memory shelf
as is illustrated in FIG. 9. Waveforms I through N and
each time a parity error is detected, counter 835 is
clocked to the next state. If during this procedure the
number of rows in which parity errors were detected
exceeds the preset state of counter 835, the OF (overflow)
output of the counter will go high placing overflow
flip-flop 836 in the SET state. -
~
-39-
~0~4ZZ4
When one complete shelf has been checked, Waveform F
returns low. In response thereto, mode circuit 816
generates the Shift and Buffer outputs as is shown in
FIG. 9. The buffer output toggles counter 824 to binary
state "01". The shift output clocks the contents of
overflow flip-flop 836 into register 837. (Waveforms B
and C also go high in conjunction with the buffer output
but have no effect at this time.) The procedure described
above then begins again. However, at this time, counter
824 is in state "01". In response thereto, selector 805
selects the four buses corresponding to shelf 1 of the
memory. Parity is then checked for each row in the second
shelf of the memory. If the number of rows containing
parity errors in this shelf exceed the predetermined
threshold number stored in counter 835, then counter 835
will again overflow, setting flip-flop 83~. At the
conclusion of the parity check for shelf 1, the buffer
and shift outputs again go high. Counter 824 is then
plac~:d in state "10" and the bit stored in
flip-flop 836 is clocked into register 837. This
procedure continues until parity has been checked for
each row in all four shelves of the memory. At the
conclusion of the parity check for shelf 3, counter 824
will be in state "11" thereby enabling gate 829
(the CKSM output is high). This in turn enables
gate 831 via gate 830. Therefore at this time when the
buffer output goes high a pulse is applied via enabled
gate 831 to the clock input of latches 838 and 839.
At this time the CKSM output is low and is applied
to the select input of latch 839. In response thereto
latch 839 selects input B applied thereto and when the
-40-
',', ~
pulse from gate 831 is applied to the clock input of
latch 839 the four bits stored in register 837 are
clocked into latch 839. These bits are then transferred
to drivers 840. Subsequent to the buffer waveform
returning low Waveform C applied to terminal 813 returns
high. This waveform is applied via gate 815 to the
EN (enable) input of drivers 840 which in response
thereto gates the 4 bits to bus 232 and from there to
the computer. From what precedes it is clear that a
bit stored in register 837 indicates that a particular
memory shelf contains a number of rows in which parity
errors have been detected, the number of rows exceeding
the threshold number used to preset counter 835.
Therefore the computer in central station 100 when
examining the 4 bits from register 837 can determine
which memory shelves have exceeded the threshold number
of rows containing parity errors. The computer can then
initiate appropriate remedial action or alternatively
commence the check sum parity check to isolate the plane
or planes which contain errors.
The computer enters the check sum mode by
applying Waveform C' to terminal 814 simultaneous with
strobing 8 bits of information into latches 810 and 811.
The 4 bits stored in latch 810 define the particular plane
in a memory section which is to be checked. These 4 bits
are applied to counter 828 and are utilized to preset
counter 828 as will be detailed hereinafter. Two of -
the 4 bits stored in latch 811 are applied to selector 806
and two of the 4 bits are applied to selector 818. In
response to Waveform C' being applied to terminal 814,
the mode circuit causes output CKSM to remain high, as
-41-
.. , ., . .. ~
10~22~
is illustrated in Waveform D' of FIG. 9 . This signal, CKSM*,
is applied to selector 806 which in response thereto
selects bus 822. The 2 bits appearing on bus 822 are
applied via selector 806 to selector 805 and function to
select four out of the 16 data buses. Similarly, the
2 bits appearing on bus 842 are applied to selector 818
and function to select one of the previously selected
four buses. Therefore, applied to retiming circuit 821,
is the data on a particular selected bus. This data has
the format indicated in Waveforms E and F of FIG. 5.
The data is retimed by circuit 821 to conform with the
timing of the parity logic and applied to input l of
gate 827. Input 2 of gate 827 is high at this time due
to the CKSM output being high. Input 3 to gate 827 goes
high each time the OF output of counter 828 goes high.
As described above, counter 828 is preset with the
4 bits stored in latch 810. These 4 bits define which
plane in a particular memory section is to be checked.
More particularly, counter 828 is preset by these
4 bits to overflow (i.e., the OF output goes high).
Counter 828 is a MOD 14 counter and therefore will
complete one cycle every 14 alock pulses. Refer to
FIG. 9, Waveform I, and assume that the data bus from
memory section 9 has been selected in the manner
described above and has been applied to input 1 of gate 827.
As is shown in FIG. 9, Waveforms F and I, the COUNT
waveform goes high in conjunction with the bit from the
first plane of a memory section. The COUNT waveform is
applied to the EN (enable) input of counter 828 such that
counter 828 will begin counting as the bit from the first
plane is applied to input 1 of gate 827. The clock
-42-
X
Z?~
associated with counter 827 is synchronized with the data
being applied to gate 827 such that counter 828 will
advance 1 state each time a bit is applied to gate 827.
Assume that counter 828 is preset to overflow with the
fourth clock pulse, i.e., counter 828 will overflow at
the time the bit from plane 3 row 0 is being applied to
input 1 of gate 827. When the OF output of counter 828
goes high gate 827 is enabled and gates the bit from
plane 3 row 0 through gate 826 and enabled gate 827
(i.e., COUNT is high) to the input of counter 835.
If this bit is a logical "1" counter 835 will advance to
a first state. Counter 828 will continue counting and
will overflow after 14 additional clock pulses.
Referring to FIG. 9, Waveform I, it can be seen that
counter 828 will overflow at the time the bit from
plane 3 row 1 is being applied to gate 827. This bit
will therefore be applied to counter 835 in the manner
described above. This process continues until all the
bit~ from the selected plane have b~ien applied to
counter 835. Counter 835 proceeds to count modulo 256
the number of binary "l's" stored in the selected memory
plane. At the conclusion of the count interval, the
COUNT waveform tWaveform F in FIG. 9) goes low. In
response thereto, the BUFFER output of circuit 816 goes
high and is applied to one input of gate 831. The
remaining input of gate 831 is also high due to the
CKSM output applied thereto via gate 830. Gate 831
therefore applies a CK (clock) pulse to the CK inputs
of latches 838 and 839. Latch 839 has selected input A
in response to the high CKSM output being applied to the
select input. Therefore latch 839 stores the 4 bits
-43-
i~\.;
lO~Z~4
from counter 835 appearing on bus 843. Latch 838 has
similarly stored the remaining 4 bits in counter 835 via
bus 844. Latches 838 and 839, in response to the buffer
pulse, transfers the 8 bits of data stored in counter 835
to the output drivers. These 8 bits of data are the
8 least significant bits of the number of "l's" stored
in the selected memory plane. Subsequent to the Buffer
output returning low, Waveforms B and C' return high.
Waveform C' is applied via path 845 and gate 815 to the
EN ~enable) inputs of drivers 840 and 844. The drivers
then gate the 8 bits of information defining the number
of binary "l's" in a particular memory plane to the
computer. It is assumed that the computer will have
stored therein the correct number of binary "1's" for
each plane and thus will be able to use the 8 bits of
information to identify particular memory planes that are
malfunctioning.
Refer to FIG. 10. Therein is shown the
details of control logic 215 which is shown in;~block
diagram form in FIG. 2. Bus 232 extending from
terminal 201 applies information to input logic 1000 and
control logic 1001. The information applied to logic 1000
consists of the ll-bit grunt address which is used to
preset the port circuits as described above. These
11 bits are inserted into logic 1000 and thereafter are
gated to memory 1006 in response to a control signal
applied to the enable input of logic 1000. Bus 232
applies a load command from the processor via path 1011
to logic 1001. Logic 1001 in response to this load
command generates certain control signals as will be
detailed hereinafter. Memory 1006 is a standard random
-44-
. ~.~ .
~O~ ~Z.~',4
access memory. This memory is utilized to store the
grunt addresses and to apply these addresses to bus 216.
Counter 1005 is a standard binary counter used to supply
addresses to memory 1006 and decoders 1007 and 1008 in a
manner to be hereinafter described. Decoders 1007 and
1008 are standard BCD/decimal decoders.
-45-
~0~42Z4
Decoders 1007 and 1008, in response to
address signals applied to the A through C inputs,
function to generate clock signals GE0 through GE6
and CE0 through CE7. The format of these clock signals
have been described above and are shown in detail in
FIG. 7. The clock signal generated by logic 202
(FIG. 2) is applied via bus 206 to input terminal 1012
and from there to counter 1005 and inverters 1009 and
1010. Inverters lOO9 and 1010 in response to the clock
signal function to provide clock signals LEA and LEB.
Clock signals LEA and LEB have been described above and
are shown in detail in FIG. 7. The grunt addresses
applied to bus 216 are in turn applied to port circuits
0 through 111 (FIG. 2) and function to preset each port
circuit in the manner described above. Clock signals
LEA and LEB, CE0 through CE7 and GE0 through GE6 are
applied to bus 217 and in turn are applied to port
circuits 0 through 111. The function of these signals
has been described above.
As described above, control logic 215 functions
to store grunt addresses received from the processor
in central station 100 and also functions to generate
the clock and control signals which have been previously
described. Logic 215 accomplishes these functions in
the following manner. Assume address counter 1005 is
initially in the CLEAR state. The processor in central
station 100 formulates the grunt addresses and
sequentially applies the addresses to bus 232 and in
turn the addresses are applied to logic 1000. Each time
a grunt address is applied to logic 1000, the processor
generates a load command and applies this command via
-46-
~094Z2~ -
path 1011 to logic 1001. In response thereto, logic
1001 places memory 1006 in the write mode by applying
a command pulse thereto via path 1004. At this same
time, logic 1001 applies a command pulse to logic 1000
via path 1003 which functions to instruct logic lOOn to
transfer the grunt on bus 232 to memory 1006. I'his
first grunt is then written into location 0 of the
memory. Logic 1001 then applies a counter enable pulse
to counter 1005 which functions to increment this counter
to the next state. The processor in central station 100
applies the next grunt address to bus 232 and generates
another load command. This second grunt is gated through
logic 1000 and stored in location 1 of memory 1006.
Counter 1005 is then incremented to the next state,
another grunt address is applied to logic 1000 and is
then written into the next location in memory 1006.
This process continues until the processor in central
station 100 has applied a grunt address for each port
card to bus 232. At the conclusion of .his process,
therefore, memory 1006 has stored in consecutive locations
a grunt address for each port circuit.
Refer now to Waveform A in FIG. 7. As described
above, Waveform A is the load port circuit control wave-
form which is applied to each port circuit via bus 217.
During the interval in which this waveform is low, the
grunt addresses are preset into the various port circuits
in the manner described above. This is accomplished as
follows. The load port circuit control waveform is applied
via path 218 to input terminal 1002 and in turn is applied
to logic 1001. In response thereto, logic 1001 applies
a reset signal to counter 1005 which resets this counter
-47-
ZZ4
to state 0. Logic 1001 also places memory 1006 in the read
mode by applying a command pulse to memory 1006 via path
1004. Counter 1005 is then allowed to begin counting
at the clock rate of clock 1012. The load port circuit
control signal is also applied to the D inputs of
decoders 1007 and 1008. A low being applied to the D
inputs enables these decoders. Counter 1005 then begins
counting at the clock rate. The output of the counter
is applied to memory 1006 and is also applied to the A
through C inputs of decoders 1007 and 1008. Counter 1005
cycles through its various states and applies sequential
addresses to the memory and to the decoders. These
sequential addresses function to read from memory 1006
the grunt addresses stored therein. These grunt addresses
are therefore sequentially read from memory 1006 and
applied to bus 216. Sequential addresses are also applied
to decoders 1007 and 1008. In response thereto, these
two decoders generate clock signals CEO through CE7 and
GEO through GE6. As these decoders are standard components,
the manner in which they generate the clock signals will
not be further described. This process continues during
the interval in which the load port circuit control signal
is low. At the end of this interval, all grunt addresses
will have been applied to the port circuits and will have
been utilized to preset these port circuits to predetermined
states in the manner described above. At the conclusion of
the load port circuit control interval, this signal returns
high. In response thereto logic 1001 applies a reset
signal to counter 1005; counter 1005 is thus reset in
preparation for subsequent lists of grunt addresses to be
stored in memory 1006.
-48-
~ .
lO~ZZ4
Although a specific embodiment of this
invention has been shown and described, it will be
understood that various modifications may be made
without departing from the spirit of this invention.
-49-
.