Note: Descriptions are shown in the official language in which they were submitted.
1156369
This is a divisional application of copending Canadian Patent
application serial NoO 336532 filed SeptO 27, 1979 filed in the names of
Araseki and Ochiai and subsequently assigned to Nippon Electric CoO, Ltd.
The present invention relates to a differential pulse-code modulation
~DPCM) srstem and, more particularly, an adaptive DPCM ~ADPCM~ system for
performing frequency band compression of speech or like signals.
The DPCM s~stem utilizing the redundancy of a speech signal is a
band compression system in which the prediction of an amplitude of each sample
of the speech signal at the present time point is made on the basis of the past
speech signal sampleO This is because the speech signal samples have a great
correlation with each otherO The simplest method of the DPCM is to give as a
predicted Yalue a sample preceding the present sample or the product of that
preceding sample and a value slightly smaller than lo The DPCM system improves
b~ about 6 dB of signal to noise ~S/N) ratio o~er the PCM system when speech
signals are transmitted with the same number of bits. In other words, the
DPCM system can save about 1 bit per sample compared with the PCM system when
the signals are transmitted with the same S/N ratioO
As a practical matter, a plurality o the past samples as well as
one past sample ma~ be used for the purpose of the band compression. In
greater detail, a predicted value ~j of a speech signal ~sample) Xj at a time
point L is gi~en by:
j-2 + A2 Xj-2 + o o o o An ~ Xj o~o~1)
~ere Al, A2, OOO~ An are called the prediction coefficients and are so
selected as to lessen the difference between Xj and~j, iOe~ a prediction
-1-
.
1 15~36g
errorO Once the optimum prediction coefficients for the speech signal are
selected, an adequate increase of n (about 5 to 8) improves S/N ratio by
approximately 10 dB compared with the PCM system. The characteristic of
the speech signal varies with time, so that the optimum values of the co-
efficients also changeO Therefore, if the optimum prediction coefficients
are seiected adaptively to the time-variation of the speech signal, the S/N
ratio can be improved by approximately 14 dB. This improvement can be
similarly achieved for other signals lying within the bandwidth of a speech
signal, such as signals given from a date modem tmodulator-demodulator)
equipment by using the DPCM system.
The prediction coefficients are obtained b~ the following two
methods: one is to analyze a speech signal for the optimum prediction co-
efficients and the other is to adaptively correct the prediction coefficients
so as to lessen the prediction error while the prediction error is being
observedO The former method must transmit the quantitized prediction error
signal and the prediction coefficients obtained. The latter method need not
transmit the prediction coefficients, resulting in simplifying the circuit
structure in the system. An ADPCM system using the latter method is discussed
by DA~ID L. COHN et alO in his paper entitled "The Residual Encoder - An
Improved ADPCM System for Speech Digitization", IEEE TRANSACTIONS ON COMMUNICA~
TIONS, VOL, COM-23, No. 9, September issue, 1975, pp. 935-9410 However since
the ADPCM system is vulnerable to transmission errors, the system needs extra
hard~are to eliminate these errors, which causes deterioration of the S/N ratio
and makes the system complicated and costly to manufactureO
Accordingly one object of the invention is to provide a transmitter
for an ADPCM system which has a simple circuit construction and which is stably
.
115~36g
operable with a great improvement of S/N ratioO
According to one aspect of the invention, there is provided a
transmitter for an adaptive differential pulse code modulation system compris-
ing: a subtractor for obtaining as output signals Ej the differences between
input signals Xj and predicted values Xj; a quantizer for quantizing the output
signals Ej from said subtractor to obtain quantized output signals Ej; and a
transmit decoder receiving said quantizer output signals and generating there-
from said predicted values, said transmit decoder comprising a predictor,
having no feedback loop, for receiving the output of said quantizer and
generating therefrom said predicted valuesO
The present invention and that of copending Canadian application
serial No. 336532 will now be described in greater detail with reference to
the accompanying drawings, in ~hich:
Figure lA is a schematic block diagram of a conventional ADPCM
system;
Figure lB is a representat~on of waveforms for describing the
system sho~n in Figure lA;
Figure 2 is a schematic block diagram of a first embodiment of the
invention; and
Figure 3 is a schematic block diagram of a second embodiment of
the invention~
In the drawings, like reference numerals represent like structural
elementsO
~3
. ~ .. . . .
,
1 15B36g
A conventional ADPC~I system will be describcd with referencc to
Fi~ures lA and lB. ~efore commencing the description, it should be under-
stood that, although the wavefor~s are expressed in analog form in Figure lB,
digital signals are used in the systems sho~ in Figures lA, ~, and 3.
Althou~h not shown, analog to digital con~erters are used at appropriate
locations, such as the preceding stage of the ADPCM systems for converting
the analog signals into digital signals.
r~eferring to Figure lA, a speech signal Xj to be transmitted is
applied to a terminal 1 of a transmitter at a time point ~. A difference
signal Ej between the input signal and the output signal Xj given from a
predictor 30 is obtained by a subtractor 10, is quantized by a quantizer 20
and is outputed from a terminal 2. The output signal Ej of the quantizer 20
and the predicted value Xj are added to each other in adder 40 and the result
of the addition is sent to the predictor 30. The predictor 30 produces the
predicted value Xj using a past input signal Xj 1 inputed to the predictor
30. The predicted value is given by:
N
Xj = ~ Ai Xj-i ''' (2)
i=l
where Ai(i = 1 to N) are prediction coefficients. The coefficients Ai are
adaptively corrected in accordance with equation (3).
i i g Fl (X~-i ) F2 (~j) ........................ (3)
where g is a positivc small value, which is about ~ 3, and Fl and F2 are
non-dccrcase functions.
The predictor 30 and tlle addcr 40 servc as a local decoder.
..~i
115B36~
A recciver, that is, decoder reccives thc signal transmitted from
the transmitter at the terminal 3. An addcr 140 calculates the sum of the
incoming signal and the output signal X; given from a predictor 1~0. The
adder 140 then prodlices a reproduction signal Xj through a terminal 4. The
decoder opcrates in the same manner as th~t of the transmitter. ~hen a
predictor 130 and the adder 140 are identical, respectively to predictor ~0
and adder 40 of the transmitter, the reproduction signal X~ in the receiver
is e~actly the same as signal X~ given from the adder 40 of the transmitter.
In this manner, without transmitting the prediction coefficients, the pre-
diction coefficients can be obtained on the basis of only the quantized pre-
diction error signal for reproduction of an original signal. The predictor
30 or 130 may be composed of the type shown in Figure 1 on page 936 in the
above-mentioned article by David L. Cohn et al.
In an actual transmission line, since a transmission error takes
place frequently, however, the above-mentioned discussion cannot be applied
to the practical system. To be more specific, the prediction errors produced
are different from each other at the transmitter and receiver and therefore
a reproduction signal is greatly different from an original signal. For the
gratual elimination of the adverse effect of the transmission error once
produced, the following equation to correct the prediction coefficients is
used:
A i = Ai (1-~) + g Fl (Xj-i) F2 ~j)
where i = 1 to N and, C is a positive value much smaller than 1, and g is
a propcr positive constant. As ~ becomcs larger, the adverse effect of the
tr~nsmission error disappears more rapidly, resulting in degracling thc
prediction perforl~ance. For e~mple, when ~ is sclectcd to be a practical
valuc, the im~rovement of S/.~ ratio is 10 dB or less. This restricts the
-5-
~.
36~
selection of thc value of ~ so as not to bc largcr in value. The constraint
of said selection allows the case where an error produced beyond the error
eliminating ability greatly degrades the speech ~uality. The most serious
problem involved in the construction shown in Figure lA is the instability
of the operation in the decoder on the receiver side having a feedback loop
when a transmission error takes place. In such a situation, since the pre-
dictor 130 and the adder 140 form a closed circuit, some of the selected
prediction coefficients might cause the receiver to oscillate or to be un-
stable in operation. In fact, it has been easily established in experiments
that intentional causing of the transmission error results in occurrence of
the oscillation or an unstable operation at the receiver. Once the operation
becomes unstable, a long time is needed until the operation settles down to
be stable. A countermeasure is taken for this problem as follows: Namely,
by monitoring the prediction coefficient on the receiving side, an unstable
operation is detected and some measure for its instability is taken on the
basis of the detection. The countermeasure, however, encounters a difficulty
in ehecking the stability of the operation, resulting inherently in the
necessity of a large scale of the system.
In the present invention, the predicted value of the speech signal
Xi is obtained from the output signal ~j of the quantizer, not from Xj 1' in
the following manner:
M
Xj = ~ Bi Ej-i
where Bji represent the prediction cocfficients in the predictors 50 and 150,
situated respectively in the transmitter and rcceiver.
This apyroach avoids thc adoption of a closed circuit in both the
~6-
115~36g
transmitter and receiver, so that even occurrencc of the transmission error
never renders its operation unstable.
Here, the coefficients Bi may be adaptively obtained by equation (5):
B ~ ) B i + g E j ~
wherein i = 1 to M, and ~ is a positive value much smaller than l to be used
to erase a detrimental effect of the transmission error. In the absence of
the quantizer 20, that is, when Ej = Ej, the transmitter serves as a filter
performing the following operation:
~1
Ej = Xj = ~ Bi Ej_i .... (6)
i=l
with the transfer function of: .
.... (7)
1 + ~ Bi Ej-i
i=l
An implementation of the just-mentioned invention, which is a first
embodiment, is illustrated in Figure 2 in block form. The transmitter
quantize~ the difference between a signal Xj and its predicted value Xj by
a quantizer 20 for transmission. The difference signal is obtained by a
substractor 10 as before. The quantizer 20 may be easily realized by utiliz-
ing techniques discussed in a paper "Adaptive Quantization in Differential
PCM Coding of Spcech" by P. Cummiskey et al., The Bell Systcm Technical
Journal, Vol. 52, No. 7, Septembcr issue, 1973, pp. 1105 to 111~. No detailed
description of thc quantizer will be given hereunder. The output signal Ej
of the quantitizer 20 is fed to a prcdictor 50. The prcdictor 50 calculates
7 _
. f ., ~......
1 15~3S~
predicted valuc ~j at a time point J by:
X. = ~ B~ E. . .... (8)
] 1 ~-1
i=l
The coefficient B~i are adaptively corrected depending on equation
~5). On the receiver side, the quanti~ed prediction error Ej is applied to
the predictor 150 which in turn produces a predicted value Xj in accordance
with equation (8). An adder 160 adds the prediction error signal Ej to the
predicted value ~j to produce a reproduction signal Xj. The reason why the
output signal Xj of the adder 160 is used as an output signal will be apparent
from the fact that if the quantizer 20 is not used, Xj = Xj. In the present
invention, when the number M of the prediction coefficients is selected to
be approximately 7 with a practical value of ~ ~i.e., about 2 6), S/N ratio
for an incoming speech signal may improve by at least 10 dB compared with the
PCM system.
Figure 3 shows a second embodiment of the invention using a pair of
predictors, which further improves the performance over the first embodiment.
In the present embodiment, the predicted value ~?j of the signal Xj is express-
ed by:
~j = Yj ~ Zj .--- (9)
where Yj and Zj are given by:
j i ~j-i .... (10)
i=l
z3 = Ai Xj-l . ... (11)
Bji arc corrected depcnding on equ~tion (53, and Ai rep~esentative
of the prediction cocfficients in thc predictors 30' and 130' are corrected
~8-
. .~ .
115~36g
dependinT on equation (12).
j+l j "~ ~
l = A- (1-8~ ~ g Ej ~j-l .... (12)
wllere ~T' = 2-3
,_
Xj in equation (11) is calucated by the adder 60 in accordance ~ith
the following e~uation (13):
Xj - Xj + Ej .... (13)
Also, on the receiver side, the Yj and Zj are calculated by the
predictors 130' and 150 and the predicted value ~j is produced from the adder
170. A reproduction signal is the output X; given from the adder 160.
In this embodiment, the predictor 130' and the adders 160 and 170
form a closed loop, so that there is a concern that the transmission error
renders the operation of the decoder unstable. Since the number of the
prediction coefficients Al in the predictor 130~ is 1, however, it is readily
seen that ¦A1 ¦ <1 is a condition for the stability of the operation.
Actually, the instability may be eliminated by adjusting both the transmitter
and receiver so as to have 0 ~ Aj ~ 0.9. In experiments carried out by
c 1
the inventors, it was observed that such adjustment provides no degradation
of the pcrformance.
In the present embodiment, if the coefficients of the predictors
50 and 150 are each 3, the S/N improvement of 14 dB is attained as compared
with the PCM system. ~hen the signal to be transmitted includes only the
speech signal, if Ajl is fixed at about 0.9, the performance never deteriorates.
The second embodiment having the two predictors appears complicated
in structure. Ilowever, if pairs of ~j 1 and Ej i~ and A3 and Bi (i = l, 2,
and 3~ are subj~ctcd to the sum of prodl~cts as is apparently understood from
equations tlO) and ~11), each opcration of the predictors 30 and 50 and thc
.
115636~
adder 70 is performcd at a time. Thc st~lcture of the embodiment shown in
Figure 3 is thcrcfore comp.~rable to that of the first embodiment shown in
Figure 2; rather, f~vorable rcsults are expected since the number of the
prediction coefficients is reduced.
l~hen it is desired to increase the number of the prediction co-
efficients in the predictors 30' and 130' of Fi~ure 3, there is no concern
that the transmission error causes the operation to be unstable, provided
each coefficient is fixed so as to stabilize the system as mentioned above.
On the other hand, when the number of the coefficients in the predictors 30'
and 130' is small, such as 1 or 2, the judgement of the stability of the
operation is performed easily despite the adaptive correction of the pre-
diction coefficients in the predictors 30' and 130'. The predictors 30' and
130' used in the embodiment of the invention have the same constructions as
those of the predictors 30 and 130 in Figure lA.
~ As mentioned above, the ADPCM system of the invention can ensure
a perfect stability of the invention, even if transmission errors take place,
with improved S/N ratio and a simple circuit construction.
_~ o_
, . . .