CA 02394981 2002-08-21
METHODS AND APPARATUS FOR DETECTING CELLULAR DIGITAL PACKET
DATA (CDPD)
This patent application is a divisional of Canadian Patent
Application No. 2,171,719, filed Septemner 14, 17~'=~
FIELD OF THE INVENTION
The present invention relates generally to methods and
apparatus for detecting a specific kind of carrier associated
with an "overlay" system.- An "overlay" system is one that
shares the frequency allocation of another system on a non-
cooperating (and non-competing) basis. one presently preferred
application of the present invention is in connection with
detecting a Cellular Digital Packet Data (CDPD) carrier, where
the CDPD system constitutes an overlay system vis-a-vis an
1~ associated cellular telephone system. This embodiment is
described herein as one exemplary application of the present
invention. However, except where they are expressly so
limited, the claims appearing at the end of this specification
are intended not to be limited to embodiments for detecting
CDPD signals.
HACRGROUND OF THE INVENTION
The CDPD system enables wireless transmission of data
over existing cellular systems, such as the Advanced Mobile _
Phone System (AMPS). The CDPD system was designed to provide -
~5 data communications in the cellular frequency range without
impeding voice communications. The system uses "mobile"
telephone channels just as a modem uses a telephone line but it
is designed to jump from one frequency to another when
. necessary, i.e., when a new telephone call starts in the cell.
The first live demonstrations of CDPD took place in
early 1993. Advocates for CDPD believe the technology will be
inexpensive and cost effective because it builds on top of
existing cellular infrastructure and does not require any
CA 02394981 2002-08-21
- 2 -
additional spectrum allocation. Instead of requiring
additional frequencies for data communications, CDPD uses
temporal "gaps" that occur between voice calls to send data in
bursts. (A voice call involves the period of time between the
dialing of a number or reception of a call to hang-up of the
handset on either terminal unit.) Technology has been
developed to "frequency hop" and seek out gaps in cellular
conversations in the cellular spectrum. The frequency-hopping
technology is expected to be sufficiently robust to handle the
designed data transfer rate (19.2 kbps) in crowded cellular
markets.
In addition, new personal wireless communications
products have been, and are being, designed to employ CDPD
technology in combining all-purpose mobile communications with
the technology of cellular phones, fax machines, modems,
electronic mail, and pen computing. Such products will provide
expanded communication capabilities to users.
Therefore, using existing hardware technology as a
building block, CDPD will allow data to be transmitted across
existing cellular networks, enabling idle voice time on
cellular systems to be filled With packets of data. CDPD
technology, with a data transmission rate of 19.2 kbps, is up
to four times faster than competing .wireless services.
However, one of the requirements of a CDPD system is the
capability of a mobile unit or base station unit to efficiently
(i.e., quickly and reliably) identify a CDPD transmission.
gU~IRY OF THE INVENTION
pn aspect of the present invention is to provide -
methods and apparatus for detecting the presence of a specific
kind of signal of an overlay system (e. g., a CDPD forward
channel) among all the channels of interest, which may be large
in number, depending on the system associated with the overlay
system. For example, there may be up to one thousand channels
of interest to an overlay CDPD system associated with a
cellular system. This specification describes a spectral
estimation technique for detection of a CDPD carrier. In
addition, this specification describes a second method that
CA 02394981 2002-08-21
- 3 -
improves on the spectral estimation method. The second method
in preferred embodiments is a minimum mean-squared error (I~iSE)
method. The spectral estimation method of the present
invention involves "shortlisting" the number of channels
scanned to a smaller number of occupied channels. In this
respect, the spectral estimation method is like another
proposed CDPD detection method (hereinafter sometimes referred
to as the "RSSI method") that uses RSSI information to
"shortlist" the number of channels scanned to a smaller number
of occupied channels. RSSI stands for Receive Signal Strength
Indicator. It is a common parameter measured in cellular and
similar communication devices to determine quality of the
received signal for the purpose of choosing the best amongst a
set of candidate channels, determining candidate channels for
handoff, or performing similar maintenance functions. It could
be used as a method to determine whether a given channel is
occupied by a signal of any kind or it is vacant. According to
the RSSI method, the mobile unit tries to identify the CDPD
channel by synchronizing and decoding each of the shortlisted
channels. One could detect the presence of a CDPD (or like)
signal on a given channel by first attempting to receive on
that channel as if indeed the signal was there (i.e.,
synchronize to the incoming signal) , collect a set of bits that
form an encoded word, attempt to decode that word, and decide
whether the signal was of the right type based on the results
of the decoding process. The problem with this approach is
that it takes a long time to synchronize to the signal, collect
enough bits to form a word and decode the word to make a -
determination about the signal. The purpose of the second
method disclosed herein is to provide a suitable alternative to
this "shortlisting" method.
The spectral estimation method detects an overlay
signal in a channel of a primary system, wherein the primary
system has a plurality of channels in which the overlay signal
might be present at any given time, and wherein the overlay
signal possesses a characteristic frequency, such as the symbol .
clock frequency. The method comprises the steps of (a)
CA 02394981 2002-08-21
-
identifying which, if any, channels) of the primary system are
occupied by a signal; and identifying which, if any, of the
signals) in the channels) identified in step (a) are an
overlay signal from the overlay system. Step (b) comprises
performing a non-linear operation on each signal to enhance a
spectral component of the signal at the carrier frequency, and
making a decision as to whether the signal is an overlay signal
on the basis of the enhanced spectral component.
In presently preferred embodiments of the invention,
the overlay signal is a cellular digital packet data (CDPD)
signal and the primary system is a cellular telephone system.
In addition, in presently preferred embodiments, step
(a) further comprises, with respect to each channel, the
substeps of (1) receiving signal and noise components in the
channel; (2) processing the received components to enhance
noise components without a prescribed spectral band relative to
noise components within the prescribed spectral band; (3)
computing the ratio of spectral components within the
prescribed spectral band to the enhanced spectral components
without the prescribed spectral band; (4) comparing the ratio
to a prescribed threshold value; and (5) identifying the
channel as carrying a signal if the ratio exceeds the
threshold.
The second inventive method disclosed herein comprises
the steps of (a) sampling signals in at least one channel to
obtain a prescribed number of samples for the at least one
channel; (b) computing an error value for each sample on the
basis of the difference, if any, between the sample value and -
the expected value; (c) determining a minimum error value; (d)
determining an average of at least two error values; (e)
determining a ratio of the minimum error value to the average;
and (f) making a decision as to whether the signal is an
overlay signal on the basis of the ratio. In presently
preferred embodiments of the second method, the error values
are mean-squared error (mse) values.
BRIEF DESCRIPTION OF T8E DRAWINGS
Figure 1 is a block diagram of the receiver of a
CA 02394981 2002-08-21
~
mobile cellular telephone unit, which is one preferred
environment for the present invention.
Figures 2A-2C illustrate a simulation employing the
inventive techniques disclosed herein. Figure 2A is a block
5 diagram of the simulator; Figures 2B, 2C and 2D are block
diagrams of a CDPD transmitter 22-1, an AMPS control channel
transmitter 22-2, and a voice channel transmitter 22-3,
respectively.
Figures 3A and 3B depict processing techniques
discussed below. Figure 3A depicts two non-linear operations,
namely, correlation 40 and squaring 42; Figure 3B depicts
filtering and thresholding employing an inner product 44,
normalization 46, and thresholding 48.
Figures 4A-4C illustrate the minimum mean-squared
error (I~iSE) processing technique, the second method disclosed
herein. Figure 4A depicts a typical demodulated signal; Figure
4B depicts a waveform and four contributions to the mean
squared error; Figure 4C schematically depicts a processor for
performing CDPD signal detection.
Figure 5 is a flowchart of the spectral estimation
detection method disclosed herein.
Figure 6 is a flowchart of the RISE detection method
disclosed herein.
DETAINED DESCRIPTION OF PREFERRED EMBODIMENTS
Figure 1 is a block diagram of the receiver of a
mobile subscriber unit (i.e., a cellular telephone capable of
receiving a CDPD signal), which represents one presently
preferred embodiment of the present invention. As shown, the
mobile unit comprises an RF (radio frequency) converter 10, IF
(intermediate frequency) ffilter 12, limiter-discriminator 14,
analog-to-digital (A/D) convertor 16, post-detection filter 18,
and identification block 20. According to the present
invention, the identification block 20 comprises means for
detecting a CDPD carrier in accordance with one of the two
methods described below. All components, with the exception of
the identification block 20, are generally well known to those
skilled in the art. Therefore, details of blocks 10-18 are
CA 02394981 2002-08-21
-s-
described herein only as necessary to explain the present
invention.
A. Method i: Spectral Estimation Detection
The spectral estimation detection method provides
signal processing solutions through manipulation of a
demodulated signal received by a mobile or base station unit. .
The present inventors have broken the detection problem into
two parts. The first part involves the detection of any signal
on the channel, and the second part involves the identification
of a CDPD signal among occupied channels. The next section
describes a signal detection technique that will identify
channels that are occupied by signals that have a prescribed
signal strength or greater. This method serves as an
alternative to scanning the RSSI over a given set of channels.
It may be performed within a time frame of 50 bit intervals
(2.6 ms) , which is comparable to that proposed for the RSSI
method.
A description of the CDPD detection problem is
provided after the description of the technique for identifying
occupied channels. Processing techniques that will yield good
discrimination between the various signals in the channel are
described. A simulation is then described and the simulation
results are tabulated.
Signal Detection (Identification of Occupied Channels)
The CDPD signal detection problem may be formulated
as a hypothesis testing problem (see, H.L. Van Trees,
Detection, Estimation and Modulation Theory Part I, Wiley,
1968):
Ho:r(t) = n(t) (la)
Hl:r(t) _ .~'(s(t) + n(t) (lb)
The received signal may be noise n(t) or a combination of
signal s(t) and noise. Since there is a possibility that the
signal is transmitted or received by a mobile unit, the
operator F represents Rayleigh fading.
The signal s(t) is a GMSK signal of the form:
CA 02394981 2002-08-21
- 7 -
s(t) _ -T cos 2nf~t + 2~ch~ a~ J ~ g(s - iT) di (2)
with the frequency pulse (see J.B. Anderson, et al., Digital
Phase Modulation, Plenum, 1986; Cellular Digital Packet Data
System Specification, Prelim. Release V. 1.0, July 19, 1993):
g( t) - 1 Q 2agT ~~T 1~2 - Q 2ngT t~= ~ ~, ( 3 )
2T n2
where T represents the symbol time (for CDPD, symbol time and
bit time are synonymous), E represents the energy per symbol
interval, h = 0.5 is the modulation index, and the product BT
is set to 0.5. The frequency pulse g(t) has been normalized so
that the phase response q (t) satisf ies conventions followed in
the representation of continuous phase modulation (CPM)
l0 signals:
( t) ° j ~ g(T) dT = 0. 5. t~~ (4)
Although the frequency pulse g(t) has infinite length,
implementations commonly truncate the GMSK filter response to
three or four symbol intervals. In addition, the non-causality
of the frequency pulse signal of equation (3) is converted into
a delay of two symbol intervals. The signal detection -
technique has been analyzed in additive White Gaussian noise
(AWGN) channels and Rayleigh faded channels, since it must work
in stationary and mobile environments.
' As mentioned above in connection with Figure l, in one
embodiment the demodulator is assumed to be a limiter
discriminator 14. The demodulated signal is modeled at
baseband, after IF filtering, as
The IF filter 12 chosen for the analysis was a four pole
CA 02394981 2002-08-21
-
QI~I~p- - $t { (I + JQ) (Q + JI) } (5)
12 + Q2 12 + Q2
Butterworth filter with cutoff at 15 kHz from the carrier.
The post-detection filter 18 in one presently
preferred embodiment is a twelve tap FIR filter at 76.8 kHz
sampling rate (corresponding to four samples per symbol
interval) with a cutoff of 6 kHz and is roughly matched to the
GMSK spectrum. The design of the post-detection filter is not
crucial to a discriminator detector implementation of the
receiver. Further details of the simulation are described
below.
The objective here is to identify a measure that
allows one to distinguish between the two hypotheses Ho, H~ .
The limiter-discriminator has the property of converting
incident white noise within the receiver bandwidth to a
parabolic spectral shape centered around the carrier. Thus,
even as the IF filter 12 suppresses incident noise at the band-
edges, the limiter-discriminator 14 will have the effect of
enhancing those spectral components. If noise alone were
passing through the limiter-discriminator, the ratio of
spectral components within the matched (post-detection) filter
bandwidth to those outside the bandwidth should be smaller than
the same~ratio calculated for hypothesis H~.
Let xk represent the output of the limiter-
discriminator 14 and let yk represent the output of the post-
detection filter (this may be, e.g., a twelve tap FIR filter
with cutoff 6 kHz}. The decision metric is then represented by
-a YN k ( 6 )
/~,~ ( x) -
~ 2
Lrk.O (XN_k _ YH_k)
where (xo, x~, .. . xN) are the first N samples available to make
a decision. The decision delay N was chosen to be fifty bit
intervals, or 2.6 ms, and yields acceptable receiver operating
characteristics.
CA 02394981 2002-08-21
g -
Identification of CDPD
The problem of identifying a CDPD signal among all
received signals may also be formulated as a hypothesis testing
problem. The hypothesis testing problem is formulated as
follows. There are two possibilities given that the channel is
definitely occupied by some signal:
Ho: The channel is occupied by CDPD (~a)
H~: The channel is not occupied by CDPD.
hypothesis H1 is in turn represented by two cases:
H~a: The channel is a voice channel (8a)
Hob: The channel is a control channel. ($b)
Typically, all Hob cases will be eliminated by the signal
detection method described above. This is because the FSK
control channel has a wider bandwidth than the CDPD signal.
Moreover, the signal will have a strong spectral component at
10 kHz, which will boost the denominator of the metric in
equation (6).
The processing techniques of the present invention ( in
applications involving detection of CDPD) rely on the fact that
CDPD uses a digital modulation technique with a fixed data
(symbol) rate of 19.2 kbps. As opposed to CDPD, voice
communications is through analog FM, and has no fixed
signalling clock other than the 6 kHz Supervisory Audio Tone
(SAT). Control channels, on the other hand, use Manchester
encoded FSK at a data rate of 10 Kbps.
The present inventors have studied two processing
techniques, aimed at locking on the 19.2 kHz symbol rate of the
output of the limiter-discriminator 14. These are
schematically depicted in Figure 3A. The first technique,
represented by circuit 40, correlates the output of the
discriminator with itself delayed by one-half bit interval; the
second technique, represented by block 42 (i.e., a square-law
device), squares the signal. It can be shown that both these
techniques produce a component at the signalling interval.
Assuming random data, whenever the CDPD signal transitions
between 0 and 1 it introduces a periodic-like segment in the
signal with a frequency of 9.6 kHz. It can be shown that
CA 02394981 2002-08-21
- 10 -
squaring a corrupted periodic waveform with a fundamental
frequency fo will produce a discrete spectral component at the
output of the square-law device at the frequency 2fo (see, W.B.
Davenport and W.L. Root, An Introduction to the Theory of
Random Signals and Noise, IEEE press, 1987). Power spectral
density calculations for the correlator 40 and square-law
device 42 have confirmed this for the CDPD signal. The output
of either non-linear device should be filtered at the frequency
of interest. A third technique involving taking the fourth
power of the matched filter output was rejected since the
effect of higher powers of a noisy signal has the effect of
noise enhancement, leading to poor performance in moderate to
low signal-to-noise environments. The simplest waveform with
a fundamental frequency of 19.2 kHz and a sampling rate of 76.8
kFiz is the sequence
cN = {1,0,-1,O,1,O,-1,0,... -1,0,1}
of length N samples. The length N was chosen to be 800
samples, corresponding to 10.41 ms of data. It should be noted
that the signal
sN = {0,-1,0,1,0,-1,0,... 1,0}
has the same fundamental frequency as, and is orthogonal to,
cN. The present invention uses the decision statistic as the
variable
2 + ~~'.o pkgkJ Z ( 9 )
where p is a vector of suitable length derived from the -
demodulated signal y processed using the correlator 40 or the
square law device 42 (Fig. 3A). Note that this is the DFT of
the processed signal at a frequency of one-quarter (1/4) the
sampling rate or 19.2 kHz if the sampling rate is 76.8 kHz.
Examination of the spectrum of the demodulated signal
subject to either of the non-linear processing methods shows
that a CDPD signal has a strong component at 19.2 kHz and an
FSK signal has a strong component at 20 kHz. This latter
component is uncomfortably close to the desired component at
CA 02394981 2002-08-21
- 11 -
19.2 kHz. In addition, since the demodulated FSK signal has
much greater power than the CDPD signal (this is due to the
wider deviation of the control signal), the output of the
filter should be normalized by the average energy in the
signal. This is done as follows: The frequency resolution of
the signal <p~ ~"> + J <P~ S"> is 76800/N. This corresponds to
96 Hz for a duration of 800 samples. If a smaller block length
is selected, say M = 40 samples, the frequency resolution would
decrease to 1920 Hz. Therefore,. a preferred normalization
technique uses the decision metric
~~ o pkcx, 2 + ~~'° pksk' 2 ( 10 )
pkCk, Z +
The effect of the above metric is essentially that of
normalizing the energy of a very narrow band filter with the
energy at the output of a filter with a broader bandwidth.
Simulation Description
The simulator implements the entire transmit chain for
all the signals, and part of the receive chain for the CDPD
mobile end-station (MES). Figure 2A is a block diagram of the
simulator. Block 22 includes. the various radio sources that
are likely to produce signals present in the channel. These
sources are a CDPD transmitter 22-1, an AMPS control channel
transmitter 22-2, and a voice transmitter 22-3. In addition,
the simulator includes a block 24 simulating Rayleigh fading,
an interpolator 26, a block 28 simulating additive white
Gaussian noise, an adder 30, a demodulator 32 (e. g., a limiter-
discriminator), and a detector 34. Blocks 32-34 of the
simulator correspond to blocks 14-20 of Figure 1.
Figures 2B, 2C and 2D, are block diagrams of the CDPD
transmitter 22-1, the AMPS control channel transmitter 22-2,
and the voice channel transmitter 22-3, respectively. As
shown, the CDPD transmitter includes a data source, phase
shaping GMSK filter, and a CPM modulator. The control channel
transmitter 22-2 includes a data source, Manchester encoder,
CA 02394981 2002-08-21
- 12 -
wideband filter, FM modulator, and 25:1 decimeter. The voice
channel transmitter 22-3 includes a white noise source,
weighting block, gain and preemphasis block, deviation limiter,
post-deviation filter, SAT source, 2 kHz deviation block,
adder, and FM modulator, configured as shown. The details of
these components are not part of the present invention and are
not described herein.
As mentioned, the receiver uses a limiter
discriminator 14 (Fig. 1) for demodulation. The front-end
filtering in the receiver is a filter derived by scanning the
front-end filter used in field trials. The limiter-
discriminator is implemented at baseband by means of a phase
detector followed by a differentiator. The phase detector
implementation requires a high sampling rate, hence the need
far the interpolator within the waveform channel. Final signal
processing is performed using samples at 76.8 kHz. This
amounts to four samples per CDPD symbol interval.
Figures 3A and' 3B depict the various processing
techniques discussed above. The two non-linear operations,
depicted in Figure 3A, are correlation 40 and squaring 42. The
filtering and thresholding are depicted in Figure 38. The
filtering and thresholding operation employs a block 44 for
obtaining an inner product, a block 46 for performing
nonaalization, and a decision block 48. Note that, in Figure
3B, the magnitude is normalized using equation (10).
An inner product is simply the sum of the element-wise
product of two vectors (or time sequences). For example, the
first term on the numerator of equation 10 is the square of the -
inner product of vectors p and cN. Normalization is the
process by which the energy is the spectral band of interest
(i.e., around 19.2 KHz, in this case) is divided by the energy
of the signal as a whole to remove gain factors that apply to
both energies and can bias the decision, if not normalized.
This process is described mathematically by equation 10.
Decision is simply the operation where the decision variable
(computed in equation l0 in this case) is compared to a
threshold and a determination is made in favor of one or the
CA 02394981 2002-08-21
- 13 -
other hypothesis depending on which side of the threshold the
decision variable lies.
Results
The inner-product technique is not computation
intensive since it implements a Fourier transfona at a fixed
frequency on a block of data. Tests have indicated that
acceptable discrimination between signals is obtainable by
processing 10.41 ms of data, corresponding to 200 bits. It is
preferable to use the signal detection method described above
to shortlist the channels. The processing described above
considers the detection and false alarm probabilities in
additive white noise and Rayleigh fading. In general, the
detection probabilities do not change much between the
correlator 40 and square-law device 42. However, the
correlator yields lower false alarm probabilities and,
therefore, it is better than the square-law device. This is
due to the fact that the square-law device implements a
memoryless operation whereas correlation has the effect of
smoothing out the effects of noise over one-half bit interval.
B. Method 2: Minimum Mean-Squared Error Detection
Identification of CDPD
As mentioned above, the CDPD problem may be formulated
as a hypothesis testing problem:
lio: The channel is occupied by CDPD, (7a)
g~: The channel is not occupied by CDPD, (fib)
The channel is a voice channel, (8a)
The channel is a control channel. (8b) -
In addition, the CDPD signal is unique in having a
fixed pulse shape and symbol time. The second inventive method
is a Least Mean Squares error approach. It may be applied
either recursively or through a block-processing technique.
simulation Description
As in the first method described above, the simulator
implements the entire transmit chain for all the signals and
part of the receive chain for a CDPD mobile end-station. The
block diagram of the simulator, along with a description of the
various radio sources that are likely to be present in the
CA 02394981 2002-08-21
- 1~ -
channel, are provided in Figures 2A-2D.
As mentioned above, the signal detection problem may
be formulated as a hypothesis testing problem stated as the
binary hypothesis test:
Ho:r(t) = n(t) (la)
Hl:r(t) _ .~"(s(t) + n(t) (lb) .'
The signal s(t) is a GMSK signal of the form:
= T cos 2nf~t + 2ah~ aj j t g(t - iT) dz (2)
s ( t)
with the frequency pulse:
g( t) - 2T Q 2nd t/T n2/2 p 2~r~f ~ t~T~ 2 , ~, (3)
where T represents the symbol time, E represents the energy per
symbol interval, h - 0.5 is the modulation index, and the
product HT is set to 0.5. The frequency pulse g(t) has been
normalized so that the phase response q(t) satisfies
conventions followed in the representation of continuous phase
modulation signals.
q( t) = j ~ g(t) dr = 0.5, t~~ (4)
Although the frequency pulse g(t) has infinite length
(support) , implementations commonly truncate the GMSK filter to
three or four symbol intervals. The non-causality of g(t) is
converted into a delay of two symbol intervals. The received
signal may be noise n (t) or a combination of signal s (t) and
noise. Since there is a possibility that the signal is
transmitted or received by a mobile unit, the operator F
represents Rayleigh fading. The technique is analyzed in
CA 02394981 2002-08-21
- 15 -
additive white Gaussian noise channels or Rayleigh faded
channels, since it should work in stationary and mobile
environments.
In one embodiment, the demodulator is assumed to be
a limiter-discriminator 14 (Fig. 1) and the demodulated signal
is modeled at baseband as in equation (5) above, with the IF
filter being a four pole Butterworth filter with cutoff at 15
kHz from the carrier. Final signal processing is performed
using samples at 76.8 kHz, which amounts to four samples per
CDPD symbol interval. The post-detection filter 18 in one
preferred embodiment is a twelve tap FIR filter at a 76.8 kHz
sampling rate with a cutof f of 9 . 6 kHz . This filter is roughly
matched to the spectrum of the GMSK frequency pulse.
Figures 4A-4C are referred to below in explaining the
MMSE processing technique. Figure 4A depicts a typical
demodulated signal, where the 10 dB point of the post-detection
filter 18 (Fig. 1) is 9.6 kHz. Figure 4B depicts a waveform
and the difference (error) between the received signal and the
expected value for the four samples in a given symbol. Figure
4C schematically depicts the processing performed by the
identification block 20 (Fig. 1). As shown in Figure 4C, the
CDPD identification block includes a T/4 demultiplexing block,
which outputs signals ak, bk, ck, dk. These signals are simply
the result of demultiplexing the stream of samples that are
obtained by sampling the analog signal at four times per symbol
period. That is, first the output of the limiter discriminator
is sampled at the rate of 4 samples per symbol period, the
samples are filtered using the post detection filter, and are
then split into four separate sets before the rest of the
processing takes place.
The signals ak, bk, ck, dk are fed to corresponding mse
(mean-squared error) computation blocks 52. The outputs of the
mse computation blocks 52 are provided to a block 54 for
determining the minimum mse and a block 54 for determining the
average of the two maximum rose's. The ratio of the minimum rose
to the average of the two maximum rose's is determined by block
5g. The output of block 58 is provided to decision block 60,
CA 02394981 2002-08-21
- 16 -
which decides which hypothesis is true, H~ or Ho. Block 58
takes as inputs the minimum Mean Square Error and the Average
of the two maximum Mean Square Errors and forms the ratio of
the two quantities. The following block (decision block 60) .
compares this ratio (also known as decision variable, denoted
by AO in block 60) to a threshold and decides in favor of a _ .
hypothesis depending on the outcome of this comparison.
The peak level of the signal at optimum timing is
assumed to be _+a. This embodiment of the invention makes a
decision on each sample of the output of the post-detection
filter 18. The decision is then used to calculate the mean-
squared error with respect to an expected signal level.
Hypothesis Ho is declared i-f the minimum of the four rose's is
below a chosen threshold. Before making a decision, the
calculated rose is normalized by the average of the two maximum
rose values. The result of this method has been compared with
the unnormalized metric, as well as another technique that used
the average of the three maximum rose values as a normalization
factor. The present method yielded faster convergence of
results than both of the others. A probable cause for the
improvement is illustrated in Figure 48. Since two samples
within a symbol interval may be close to the optimum sampling
point, the contributions to the mean-squared error due to those
two points may be close. Thus, inclusion of one .of those
points in the normalization factor adversely affects the
discrimination ability of the method.
Results
Results have been obtained for processing intervals
of 2.6 ms and 5.2 ms. The longer processing interval could be
used for confirmation of coarse results from a quick pass over
all channels of interest with a smaller processing interval.
However, there is seen to be a marginal improvement in results
at low signal-to-noise ratios, and it may be worth scanning
through the whole set again using 2.6 ms of data. A maximum of
three passes will increase the probability of detection from
95% to 99.98%. Assume the total number of channels that the .
mobile end station scans over is 1000. One pass will take a
CA 02394981 2002-08-21
- 17 -
maximum of
2.6 s. Table
1 gives
a list of
probabilities
for
interesting
events.
A nominal
probability
of detection
for one
pass over
the set
of channels
is chosen
to be 0.95.
The
simplified
scenario
in the table
does not
consider
the
possibility implemented
of false-alarm. scheme
The would
preferably of
make a pass channels,
over the short-listing
set
candidates pass
in the process. over
A second the
short-listed
candidates more
is not expected than
to add 1
second
to
the
processing
time, under
worst case
channel
conditions.
EVENT PROBABILITY
One pass succeeds 0.9500
One of three passes succeeds 0.9998
Detection time < 2.6 s 0.9500
Detection time < 5.2 s 0.9975
Detection time < 7.8 s 0.9998
Table 1: A list of interesting events and
corresponding probabilities of detection.
The expected signal value is dependent on the
implementation of the limiter-discriminator 14 (Fig. 1). It is
recommended that automatic calibration of the scheme be
implemented. This calibration could be dynamically done. The
mobile end-station (or base station) could recalibrate whenever
the system registers on a CDPD channel. -
There are two parameters that should be optimized for
fine tuning the signal processor. One of them is the bandwidth
of the post-detection filter 18 (Fig. 1) . This need not be
identical to the filter actually used during reception. A
wider bandwidth will yield less intersymbol interference but
~ will allow more noise at the filter's output. Since, however,
the expected signal value in no noise conditions depends on the
- amount of intersymbol interference, a slightly wider bandwidth
will probably be more effective. The other parameter that
CA 02394981 2002-08-21
- 18 -
needs tuning is the expected signal level ~a. The optimum
value of a will be the average peak sample value at the 76.8
kHz rate. For the particular implementation disclosed herein,
the value chosen was 0.6. This value will change depending on
the implementation.
The threshold choice for a first pass should allow a
20-30% false alarm probability. Such a choice will yield
detection probabilities in excess of 98% at all SNR's of
interest. On the second pass, the false alarm rate may be
1D greatly reduced.
The two inventive methods disclosed herein will now
be summarized with reference to Figures 5 and 6.
Referring first to Figure 5, the Spectral Estimation
Detection Method begins by computing energy within the filter
band width. The energy outside the filter band width is then
computed. Next, the decision variable is formed, for example,
in accordance with Equation 6. The decision variable is then
compared with a threshold. If the decision variable is not
greater than the threshold, it is determined that no signal is
present. On the other hand, if the decision variable is
greater than the threshold, the method declares that a signal
is present. Thereafter, the signal is passed through a
correlator or a square-law device (for example, the.correlator
40 or square-law device 42 of Figure 3A. Thereafter, a
decision variable is formed in accordance with Equation 10. A
determination is then made whether the decision variable is
greater than a threshold. If not, the signal on the channel is
declared a non-CDPD signal. If the decision variable is -
greater than the threshold, the signal is declared a CDPD
signal. All channels are preferably scanned in this manner.
Referring now to Figure 6, the MSE method begins by
demultiplexing the signal output of the post-detection filter
18 (see Figure 1). Thereafter, the MSE of each demultiplexer
branch (see Figure 4C) is computed. The minimum MSE is then
3~ found, and then the average of the two maximum MSEs is found.
The ratio of the minimum MSE to the average of the two maximum
MSEs is then compared to a threshold. If the ratio is not
CA 02394981 2002-08-21
- 19 -
greater than the threshold, the signal is declared a non-CDPD
signal. If the ratio is greater than the threshold, the signal
is declared a CDPD signal. Finally, as with the Spectral
Estimation Method, all channels are preferably scanned.
The present invention may be embodied in other
specific forms without departing from the spirit or essential
attributes thereof, and, accordingly, reference should be made
to the appended claims, rather than to the foregoing
specification, as indicating the scope of the invention and the
scope of protection of the following claims.
