 |
Description  |
|
|
FIELD OF ART
The invention relates to a receiver for a DSSS signal (DSSS=direct sequence
spread spectrum) wherein digital data in the form of symbols are spread
out in the DSSS signal with a predetermined pulse sequence having the
length L.
STATE OF THE ART
Efforts have been underway for some time to realize a digital mobile radio
network. Corresponding standards have already been defined as well (for
example, GSM: Narrow-Band Pan-European Mobile Radio System; CD 900: Civil
Band-Spread System). However, development has nowise reached its end.
There is still the question how to optimally utilize the terrestrial
mobile radio channel, eminently poorly suited for the transmission of
digital data, in principle. There is not only the objective of
eliminating, in the best possible way, the occurring disturbances
(multipath propagation, time- and frequency-dependent fading, etc.), but
also the goal of efficiently realizing this feature from the viewpoints of
circuitry and bandwidth.
Band-spread systems tailored a priori to multipath propagation of the
mobile wireless transmission channel are relatively immune to jamming but
exhibit the drawback that their implementation is expensive. For these
reasons the band-spread technique has been restricted primarily to
military applications.
The principle in the transmission of a so-called direct sequence spread
spectrum signal, called briefly DSSS signal hereinbelow, resides in that
the data to be transmitted are multiplied with a predetermined pulse
sequence having the length L. In this process, the pulse sequence has a
chip rate amounting to L-fold the symbol rate. In the receiver, the DSSS
signal is detected with a so-called matched filter or with a correlator
(compare "Principles of Communication Systems", W. Taub, D. L. Schilling,
McGraw Hill, Singapore 1986, pp. 720-727).
An analysis of the advantages and drawbacks of DSSS systems is provided by
the article "Spread Spectrum Communications - Myths and Realities", A. J.
Viterbi, IEEE Communications Magazine, May 1979, pp. 11-18.
In order to be able to create a reliable data transmission, it is important
to determine the channel transient response. In mobile radio systems where
the participants are constantly in motion, it is, of course, impossible to
measure the channel once and for all and to set the filter correspondingly
in the receiver. Rather, a method must be devised which determines the
channel transient response continuously or at regular time intervals.
In this connection, attention is to be invited to the dissertation by Jurg
Ruprecht, "Maximum-Likelihood Estimation of Multipath Channels", Diss. ETH
No. 8789, Zuerich, 1989, Hartung-Gorre publishers. The task of channel
estimation, important for mobile radio, is attained by transmitting a
known pulse sequence having well-defined properties via the channel and
filtering same in the receiver by a filter inverse with respect to the
pulse sequence The inverse filter here reacts to the transmitted pulse
sequence in such a way that the desired pulse response of the channel
appears at the output of the filter.
SUMMARY OF THE INVENTION
It is an object of the invention to provide a receiver of the type
discussed above suitable, in particular, for receiving digital data on
mobile wireless transmission channels and lending itself to being
efficiently implemented from the viewpoint of circuit technology.
According to the invention, this object has been attained by providing, for
detecting the symbols, an approximated inverse filter which, upon
excitation by the predetermined pulse sequence (s[.]) as such responds
approximately with a Kronecker delta sequence as the output sequence.
The basic aspect of the invention resides in that, for detecting the data,
the DSSS signal is processed by means of an inverse filter rather than
with a matched filter. The matched filter, though being the better filter
with respect to maximum signal-to-noise ratio, is surpassed by factors by
the inverse filter with regard to the so-called peak/off peak ratio (ratio
of maximum amplitude to largest spurious amplitude of the filtered DSSS
signal). The pulse peaks at the filter output, containing the transmitted
data, thus can be detected in a simpler and more reliable fashion.
There is a substantial difference between the transmission of data and the
channel estimation by means of known pulse sequences. In channel
estimation, the receiver knows the transmitted "data", namely the
individual chips of the pulse sequence, beforehand with regard to amount
and arithmetic sign, whereas in data detection unknown data (symbols) with
which the pulse sequence is multiplied must be determined. The pulse
sequence utilized in the DSSS signal is known as such to the receiver, but
this is not true for the symbol with which the sequence has been
multiplied in the concrete instance.
A substantial advantage of the invention resides in that the channel
estimation is integrated into the data detection, for it constitutes part
thereof. Correspondingly, the circuitry required for the receiver side is
smaller than in the conventional matched filter receivers.
The ideal inverse filter has infinitely many coefficients. It reacts to the
pulse sequence as such with a pure Kronecker delta sequence, i.e. the
digital system response is different from zero only at the instant i=0.
(With a filter length of L+2M, this corresponds to the borderline case
M.fwdarw..infin..)
Under practical conditions, the inverse filter can, of course, be realized
only in an approximation. However, because the coefficients v[k] of the
ideal inverse filter, in case of a suitably selected pulse sequence, will
disintegrate exponentially for large k, rather satisfactory results can be
achieved already with filters of reasonable length. In accordance with an
advantageous embodiment, the coefficients v[k] of the inverse filter obey
the following relationship:
##EQU1##
For -M.sub.1 .ltoreq.k.ltoreq.L+M.sub.2, the filter coefficients correspond
to the ideal case; outside of the mentioned range they are simply set at
zero. The approximated inverse filter therefore corresponds to the
truncated ideal inverse filter (M.sub.1 and M.sub.2 are predeterminable
numbers, i.e. so-called design parameters).
An alternative resides in limiting with respect to time and advantageously
weighting the coefficients by multiplication with a suitable window
function.
There are various approximation strategies. Besides truncation of the ideal
inverse filter, the following two approximation solutions are of
importance, in particular:
1. Least square approximation;
2. Approximation for maximum POP ratio.
In the first case, the approximated inverse filter responds upon excitation
by the predetermined pulse sequence (s[.]) as such with an output sequence
approximating the Kronecker delta sequence along the lines of minimum
errors squared.
In the second case, the output sequence has a maximum peak/off peak ratio.
In other words, the largest secondary peak is made as small as possible as
related to the main peak. This involves an approximation to the infinitely
large peak/off peak ratio of a Kronecker delta sequence.
The pulse sequence is preferably chosen so that it results in a maximum
process gain G. In this connection, process gain is understood to mean the
inverse of the energy of the inverse filter.
##EQU2##
Thus, the process gain should come maximally close to the matched filter
bound, i.e. the theoretically determined upper limit. The matched filter
bound corresponds to the sequence length L, i.e. G.ltoreq.L.
In other words The process gain G of the selected pulse sequence should be
maximal with respect to as many as possible, especially all, pulse
sequences of a given length L.
Maximizing of the process gain leads to an MMSE estimation of the channel
(minimum mean square error).
After the inverse filter, a so-called "matched filter detector" called, in
short, MF detector, is preferably connected. This is a circuit arrangement
comprising essentially a matched filter adapted to the channel and a
threshold value detector The MF detector provides, in first approximation,
a maximum likelihood detection of the data symbols.
By the use of the inverse filter, channel estimation is delivered
concomitantly with the data detection almost automatically. It is
recommended, along the lines of a decision return (decision feedback),
with the aid of the data estimated in the receiver to cancel the
multiplicative effect of the symbols on the output signal of the inverse
filter (reduction of the output signal), to average the channel transient
responses obtained in the various symbol intervals, and to use same for
setting the coefficients of the channel-adapted matched filter.
In order to be able to estimate the channel transient response in a
maximally distortion-free fashion, the symbols should have a symbol period
larger than the duration of the channel transient response.
In view of the signal amplification in the transmitter, it is advantageous
for the pulse sequence to have a constant envelope curve. This means
nothing else but that the individual pulses of the sequence are all
identical with regard to their amount.
The invention can be realized in a particularly simple way with binary
pulse sequences and, selectively, also with binary symbols. The reduction
of the output signal mentioned in connection with the channel estimation
then requires essentially a simple multiplication.
A typical field of use for the invention is the mobile radio technology, be
it in the macro- or in the microcellular region. The advantages of the
DSSS signals in conjunction with the inverse filter will become apparent,
above all, in so-called in-house applications (microcellular mode) since
here the signal transit times and the duration of the pulse responses are
relatively small (typically<500 ns).
From the totality of the dependent claims, additional advantageous
embodiments can be derived.
BRIEF DESCRIPTION OF THE DRAWINGS
The invention will be described in greater detail below with reference to
embodiments and in conjunction with the drawings wherein:
FIG. 1 is a schematic view of the transmitter and of the receiver;
FIG. 2 shows the functional principle of the ideal inverse filter;
FIG. 3 shows a schematic view of the inverse filter;
FIGS. 4a, and 4b show schematic views of a window function for the inverse
filter;
FIG. 5 shows the functional principle of the real inverse filter;
FIG. 6 is a block circuit diagram of an MF detector with decision feedback;
and
FIGS. 7a, and 7b show the principle of channel estimation.
The reference numerals utilized in the drawings and their meanings are
summarized in the list of symbols. Basically, identical parts in the
figures bear the same reference symbols.
WAYS OF EXECUTING THE INVENTION
FIG. 1 shows the illustration, in principle, of a transmitter/receiver
structure according to this invention. A transmitter 1 which, for example,
is part of a radiotelephone transmits digital data (e.g. a digitalized
voice signal) via a channel 2 to a receiver 3, for example to a base
station of the mobile radio system. The channel 2 has a time-variant
transient response h(t) typically characterized by a multipath
propagation. The transmission signal is furthermore heterodyned by
additive white Gaussian noise W(t).
The digital data are present in the transmitter in the form of symbols
B.sub.m at a given time interval T.sub.s (symbol period). In a spreading
circuit 4, the symbol period T.sub.s is divided into a predetermined
number of L chip intervals T.sub.c (chip period), i.e. T.sub.s =LT.sub.c
The output signal is identical to zero in all chip intervals save one. For
the sake of simplicity, it is assumed that this is the first interval.
A pulse sequence generator 5 produces, from the chronologically expanded
symbols, a DSSS signal of the form:
##EQU3##
The DSSS signal U[i] thus is composed of time segments having the length
LT.sub.s wherein there are accommodated in each case the L pulses of the
pulse sequence s[.], multiplied with the symbol B.sub.m.
The DSSS signal U[i] is converted in a subsequently connected pulse
modulator 6 and a low-pass filter 7 into a time-continuous signal, limited
with respect to frequency, which is then modulated in a conventional
modulator 8 onto a carrier oscillation in a manner known per se.
The transmitted DSSS signal is first demodulated from the carrier
oscillation in the receiver 3 (demodulator 9), freed of undesirable
frequency components in a low-pass filter 10, and thereafter scanned in
correspondence with the chip rate 1/T.sub.c (scanner 11).
In accordance with the basic aspect of the invention, the now present
received signal is processed through an inverse filter 12. The output
signal Y[i] of the inverse filter is finally evaluated by an ML (maximum
likelihood) detector 13 so that lastly the estimated symbols (B.sub.m
characterized by a circumflex) are present.
The properties and effects of this central inverse filter 12 will be
described in detail below.
FIG. 2 illustrates the effect of the inverse filter 12. It is tuned in a
quite specific way to the pulse sequence s[.] utilized by the transmitter.
The left-hand half of the figure shows, as an example, an aperiodic pulse
sequence s[.]={s.sub.0, . . . , s.sub.L } of the length L=13 (s[i]=0 for
i<0 and i>L). When introducing the sequence as such into the corresponding
inverse filter v[.], then, in the ideal case, there results as the system
response a Kronecker delta sequence .delta.[k] (see right-hand half of the
figure). Such a sequence is distinguished, as is known, by the fact that
it is different from zero only at the instant i=0:
##EQU4##
The ideal inverse filter has an infinite number of filter coefficients
v[k]. Under practical conditions, such a filter cannot, of course, be
realized exactly. For this reason, the inverse filter is limited to a
length of -M.sub.1 .ltoreq.k.ltoreq.M.sub.2 +L. Outside of this window,
the filter coefficients v[k] are identical to zero. The L+M.sub.1 +M.sub.2
filter coefficients are fixed so that the output signal of the inverse
filter resulting upon excitation with the pulse sequence s[.] approximates
maximally well the Kronecker delta sequence .delta.[k]. The type of
approximation is defined with a suitable mathematical criterion. Three
advantageous approximation strategies shall be mentioned below.
The filter coefficients v[k] different from zero are chosen, according to a
first embodiment of the invention, so that the system response to the
corresponding pulse frequency corresponds to a Kronecker delta sequence
only within a predetermined time window. In other words, within the time
window having the length -M.sub.1 . . . +M.sub.2, the system response is
different from zero only at the chronological zero point (i=0). Outside of
the aforementioned window, in contrast thereto, there will definitely
occur secondary peaks.
FIG. 3 shows schematically the coefficients of an inverse filter. L denotes
the length of the corresponding pulse sequence. Based on L, the inverse
filter is lengthened on both sides by a predetermined number M.sub.1 and,
respectively, M.sub.2, of coefficients. Thus, in total, it has L+M.sub.1
+M.sub.2 filter coefficients. In accordance with a preferred embodiment,
they obey the following rule:
##EQU5##
For -M.sub.1 .ltoreq.k.ltoreq.L+M.sub.2, the filter coefficients correspond
to the ideal coefficients; outside of the mentioned range they are simply
set at zero (truncation). A thus-designed filter has been known as such
from the above-mentioned publication by J. Ruprecht. It was also
demonstrated in the latter that the coefficients (with suitable choice of
the pulse sequence) will disintegrate exponentially for large k and
consequently the truncation will yield definitely usable results.
An alternative to truncation is represented by weighting with a window
function which, at the margin, passes gently rather than abruptly toward
zero.
FIG. 4 illustrates the two variants. In FIG. 4a, the truncation (constant
weighting within the window) is shown, and FIG. 4b illustrates a window
function with weighting that levels off at the margin.
FIG. 5 shows the effect of truncation on the system response of the inverse
filter when introducing the corresponding pulse sequence as such. A single
peak occurs within the time window -M.sub.1 . . . +M.sub.2. Outside
thereof, several secondary peaks occur (in principle, an infinite number).
However, these are very small and can be covered by means of the POP
ratio. The POP ratio is defined as the ratio between the amount of the
amplitude of the main pulse A.sub.0 and that of the largest secondary
pulse A.sub.max.
Besides the just-described strategies, there are still two more preferred
approximation strategies:
1. Least square
2. Approximation for maximum POP ratio.
In the first case, the approximated inverse filter, upon excitation by the
preset pulse sequence s[.] as such, responds with an output sequence
approximating the Kronecker delta sequence .delta.[k] in the sense of the
smallest square errors:
##EQU6##
In the second case, the output sequence has a maximum peak/off peak ratio.
In other words, the largest secondary peak is made as small as possible as
compared with the main peak. Here an approximation is involved to the
infinitely large POP ratio of a Kronecker delta sequence.
An important property of the inverse filter resides in that these secondary
peaks, which can be quantitatively covered mathematically with the
peak/off peak ratio (POP ratio), can be kept very small with a suitable
choice of the pulse sequence.
A central advantage of the inverse filter thus resides in that its POP
ratio is much larger than that of a matched filter (respectively
correlator). Whereas the POP ratio in a matched filter typically ranges at
about 10, even a poor inverse filter realizes POP ratios of 100 and more.
With optimization, it is normally possible to easily attain 30-40 dB. The
inverse filter, in this aspect, is thus superior by orders of magnitude to
the matched filter. It is interesting to note that this advantage need not
be obtained at the cost of a correspondingly grave impairment of the
signal-to-noise ratio.
Preferably, the pulse sequence s[.] is designed so that the inverse filter
exhibits a maximum process gain G (process gain). Process gain in this
connection is understood to mean the inverse of the energy of the inverse
filter:
##EQU7##
If, for one L, there exist several pulse sequences with the same G, then it
is recommended (with a given M.sub.1 and M.sub.2) to select the one having
the largest POP ratio. Examples of such so-called optimal sequences ("best
invertible sequences") can be derived from the dissertation by J.
Ruprecht. However, for practical applications, lengths of below L=10 are
meaningless. Rather, relatively large lengths are desirable (for example L
going toward 100).
For large lengths (e.g. L>100), it is difficult, if not impossible (since
the calculating time for calculating 2.sup.L possibilities if frequently
too long) to find the pulse sequence having the maximum process gain with
respect to all sequences. Therefore, under practical conditions, a group
of L sequences with tendentially good properties will be chosen, and the
process gain will be optimized with respect to this limited group. The
sought-for process gain will then represent a relative maximum.
With a view toward signal amplification in the transmitter, it is
advantageous for the pulse sequence to have a constant envelope curve.
This means nothing else but that the individual pulses of the sequence are
all of the same size with regard to their amount. Preferably, a binary
sequence is involved (i.e. s[i]=+/-1). The maximum transmission power then
corresponds precisely to the mean power.
A great advantage of the invention resides in that the channel estimation
is concomitantly provided almost without any additional effort as a
"by-product" of the data detection. The DSSS signal transformed during
transmission over the channel namely does not produce at the output of the
inverse filter a pure Kronecker delta sequence but rather yields the
(equivalent, time-discrete) transmission function h[.] of the channel,
multiplied by the symbol value B.sub.m. The problem, now, resides in that
the receiver does not know the transmitted symbols. The following
description will explain how the data detection and the channel estimation
are performed.
FIG. 6 shows in detail the functional blocks contained by the ML (maximum
likelihood) detector 13 illustrated in FIG. 1. In principle, two tasks are
accomplished: Firstly, the symbol detection according to the ML principle
and, secondly, the channel estimation along the lines of a decision
feedback.
The symbol detection proceeds in accordance with a process known per se.
First of all, the output signal Y[i] is filtered by a matched filter 14
with the transient response h*[-.] ("*" denotes the conjugated-complex
value; "-" denotes the time reversal). The filter 14 is thus adapted to
the time-discrete channel h[.] equivalent to channel 2. Thereafter, in a
scanning circuit 15, the correlation maximum is scanned and thus the clock
rate is reduced by the factor L. Thus, a changeover is made again from the
chip interval T.sub.c to the symbol period T.sub.s.
In order to obtain the estimated symbol values B.sub.m, it is sufficient to
reduce the complex-value scanned output signal of the matched filter 14 to
the real portion (real portion extractor 16) and to effect discrimination
with a threshold value detector 17.
For the channel estimation, the output signal Y[i] of the inverse filter is
converted in a serial/parallel converter 19. The signal values Y[i]
scanned with the chip period T.sub.c are converted into vectors Y.sub.m of
the dimension L. Each vector Y.sub.m contains the scanning values lying
within the same symbol interval m. Each vector Y.sub.m corresponds to a
sample function of the channel transient response multiplied by an unknown
symbol value.
The symbols estimated in the data detection path are then utilized for
"reducing" the aforementioned vectors Y.sub.m, i.e. for eliminating the
multiplicative "symbol proportion" (in principle by means of a division).
In the preferred binary case (B.sub.m =+/-1), the reduction consists
essentially in a multiplication 18 (since s.sub.i.sup.2 =1 applies).
The thus-reduced vectors Y.sub.m ' m are fed into a channel estimator 20.
According to a preferred embodiment, the latter performs averaging over
various vectors (averaging over several realizations of the channel
transient response which is burdened by noise and changes gradually with
time):
##EQU8##
Averaging is realized, for example, with the aid of a delay member 21 and a
feedback path. The type of averaging is determined by the choice of the
weighting factors w.sub.i. The weighting factors wi proper can be fixed
according to conventional principles. Of course, it is also possible to
utilize more sophisticated methods for channel estimation.
The filter coefficients of the matched filter 14 are the time-inverted and
conjugated-complex components of the averaged vectors.
FIGS. 7a, b are to illustrate the procedure in channel estimation for the
binary case. In each case, the time is plotted on the abscissa and the
signal amplitude on the ordinate. FIG. 7a shows the output signal Y[i] of
the inverse filter. Respectively L scanning values are combined into a
vector Y.sub.m. In the present case, the vectors differ essentially only
by the polarity (according to the illustration, the following applies
regarding the present example: B.sub.m =+1, B.sub.m+1 =-1, B.sub.m+2 =-1).
The individual vectors Y.sub.m are then reduced by multiplying with the
estimated symbol value
##EQU9##
Thus, the influence of the unknown data is eliminated; all vectors have
the same polarity (FIG. 7b). For the channel estimation, averaging in
component fashion is then carried out, as mentioned above.
For a good channel estimation, the symbol period should always be so long
that it is larger than the duration of the channel transient response.
This ensures that the individual sample functions can be readily
separated, i.e. that neighboring transient responses will not overlap.
The prerequisites, in principle, for determining an individual sample
function of the channel transient response with the aid of an aperiodic
pulse sequence and an inverse filter are described in the cited
publication by J. Ruprecht. They can be applied analogously to the channel
estimation according to this invention.
Maximizing of the process gain leads to an MMSE estimation of the channel
(minimum mean square error). For a good performance, it is recommended to
select long sequences with a high process gain G. The maximally realizable
process gain, however, is for fundamental reasons never larger than L
(so-called "matched filter bound").
As a final remark, it can be noted that a receiver has been created by the
invention which utilizes the advantages of the band-spread technique in an
efficient way for mobile radio transmission.
List of Symbols
1--transmitter
2--channel
3--receiver
4--spread circuit
5--pulse sequence generator
6--pulse modulator
7--low-pass filter
8--modulator
9--demodulator
10--low-pass filter
11--scanner
12--inverse filter
13--ML detector
14--matched filter
15--scanning circuit
16--real portion extractor
17--threshold value detector
18--multiplication
19--serial/parallel converter
20--channel estimator
21--delay member
B.sub.m --symbol
B.sub.m.sup.circumflex --estimated symbol
* * * * *
|
|
 |
|
 |
Description |
|
