METHOD AND APPARATUS FOR SOFT INFORMATION GENERATION IN JOINT DEMODULATION OF CO-CHANNEL SIGNALS
BACKGROUND OF THE INVENTION
The present invention is directed toward wireless communications systems, and, more particularly, to an apparatus and method for soft information generation in joint
demodulation of co-channel signals. The performance of receivers in wireless communications systems, such as mobile communications systems, may degrade severely due to multipath fading. Although
anti-fading techniques, such as antenna diversity, equalization and adaptive ray processing, are effective in improving the performance of the receiver, forward error correction (FEC) techniques may be necessary to achieve acceptable voice and data transmission in wireless communications systems. FEC techniques provide redundancy by adding extra bits to the actual information bits, which allows the decoder to detect and correct errors. To optimize
decoder performance, it is important to provide accurate soft information from the
demodulation process. The availability of soft information, after the demodulator, increases the performance of the concatenated (demodulator and decoder) structure considerably when compared to the use of hard information.
The generation of soft information has been used extensively in conventional
single user demodulators. Several approaches for developing soft information in single user
demodulators, such as soft information for maximum likelihood sequence estimation (MLSE)
detection for inter symbol interference (ISI) channels, have been presented. These techniques
have been extended to π/4 shifted-DQPSK (differential quadrature phase shift keying)
systems.
In code-division multiple access (CDMA) systems multiuser demodulation has
gained attention. Multiuser demodulation has only recently been adopted in narrow band
time-division multiple access (TDMA) base systems. Soft information generation for multiuser demodulation in CDMA is discussed in S. Verdu, "Multiuser Detection",
Cambridge University Press, 1998. However, multiuser demodulation in TDMA is different from CDMA. Moreover, soft information generation for π/4-DQPSK is different from soft information generation for BPSK (bit phase shift keying) and QPSK.
SUMMARY OF THE INVENTION
The present invention relates to a method and apparatus for soft information
generation injoint demodulation of co-channel signals, particularly π/4-DQPSK co-channel signals.
Particularly, the method and apparatus determines jointly detected symbols and corresponding joint metrics. A first potential nondetected metric sum is determined by flipping the current desired coherent detected symbol and keeping the interfered coherent
detected symbol as is. A second potential nondetected metric sum is determined by flipping
the previous desired coherent detected symbol and keeping the interfered coherent detected
symbol as is. A minimum between the difference of the detected joint metrics and the two
nondetected joint metrics is found. The soft values are then established based on the
minimum difference metric.
Further features and advantages of the invention will be apparent from the
specification and the drawings.
BRIEF DESCRIPTION OF THE DRAWINGS
Fig. 1 is a block diagram of a mobile communication system using the method and apparatus in accordance with the invention;
Fig. 2 is a block diagram of the joint demodulator of Fig. 1 for a first
embodiment of the invention;
Fig. 3 is a block diagram of the joint demodulator of Fig. 1 for a second embodiment of the invention;
Fig. 4 is a flow diagram illustrating a program implemented in the joint demodulator of Fig. 1; o
Fig. 5 is a trellis diagram illustrating soft information generation for joint
demodulation for flat fading channels in accordance with one aspect of the invention;
Fig. 6 is a trellis diagram illustrating soft information generation for joint
demodulation for a dispersive channel in accordance with a first aspect of the invention;
'Fig. 7 is a trellis diagram illustrating soft information generation for joint
demodulation for dispersive channels in accordance with a second aspect of the invention;
Fig. 8 is a trellis diagram illustrating soft information generation for joint
demodulation for dispersive, channels in accordance with a third aspect of the invention; and
Fig. 9 is a trellis diagram illustrating soft information generation for joint
demodulation for a dispersive channel in accordance with a fourth aspect of the invention.
DETAILED DESCRIPTION OF THE INVENTION
Fig. 1 is a block diagram of a typical mobile communication system 10, such as IS-136, using π/4 shifted-DQPSK (differential quadrature phase shift keying). The mobile communication system 10 includes a first transmitter 12 having' a first transmit antenna 14. A second transmitter 16 has a second transmit antenna 18. A receiver 20
includes a receiver antenna 22. For simplicity, only two transmitters and one receiver are shown in Fig. 1. However, the proposed invention can be applied to more than two transmitters and a receiver. In the illustrated embodiment of the invention, the receiver 20 is described as the receiver of a mobile terminal while the transmitters 12 and 16 are
associated with respective base stations as part of fixed terminals, as is known. Alternatively,
the receiver 20 could be the receiver in. a base station, while the transmitters 12 and 16 could
be the transmitters in mobile terminals, or any combination thereof.
The present invention is described herein in the context of a mobile terminal.
As used herein, the term "mobile terminal" may include a mobile communications radio
telephone with or without a multi-line display; a personal communications system (PCS) terminal that may combine a mobile communications radio telephone with data processing,
facsimile and data communications capability; a PDA that can include a radio telephone, pager, Internet-Intranet access, Web browser, organizer, calendar and/or a global positioning
system (GPS) receiver; and a conventional laptop and/or Palm© top receiver or other appliance that includes a radio telephone transceiver. Mobile terminals may also be referred
to as "pervasive computing" devices.
Each transmitter 12 and 16 includes an encoder 13 and a differential modulator 15. Information bits are encoded in the encoder 13 and the differential modulator
15 modulates theses encoded bits using π/4 shifted-DQPSK.
The invention is described under the assumption that the first transmitter 12 transmits desired information signals, and the other transmitters, such as the transmitter 16, transmit interfering signals, also referred to "interferers". The receiver 20 attempts to receive
the desired user information signals correctly under the presence of the interferers and thermal noise, represented at 28, Both transmitted signals reach the receiver 20 after passing
through independent propagation mediums (e. g., mobile radio channels represented at 24 and 26). The transmitted signals plus the noise 28 are received at the receiver antenna 22,
While a single receiver antenna 22 is shown, the receiver 20 could have more than one antenna. The received signal is processed by a radio processor 30 which amplifies, mixes,
filters, samples and quantizes the received signal to produce a baseband signal. The baseband signal is supplied to a joint demodulator 32 in accordance with the invention. The joint
demodulator produces soft values which are supplied to a decoder 34. As mentioned above, channel encoding is frequently used in transmitters to provide redundancy by adding extra
bits to the actual information bits. In the receiver 20, the decoder 34 is used to decode the
encoded bits while detecting and correcting possible errors in the received signal.
While a specific receiver block diagram is provided herein for the. purpose of illustration, those skilled in the art will appreciate that other known architectures may also be used. Additional blocks, such as interleaving, and the like, are not mentioned herein for
purposes of simplicity.
In accordance with the invention, a method and apparatus generates soft
information in joint demodulation of co-channel signals. Particularly, soft information values are generated from signal samples of differentially encoded signals by demodulating both desired and interfering signals together. However, soft information values are generated for only the desired signal.
Referring to Fig. 2, a block diagram of the joint demodulator 32 of Fig. 1 , in
which soft information generation in joint demodulation of co-channel signals can be applied,
is illustrated. The baseband received signal is applied to a likelihood information generator
36 and to a channel estimator 38. An output of the channel estimator 38 is also applied to the likelihood information generator 36. The output of the likelihood information generator
36 is applied to a likelihood information processor 40. The likelihood information processor 40 develops the soft values supplied to the decoder 34 of Fig. 1.
The channel estimator 38 estimates amplitude and phase information
corresponding to the mobile radio channels for both the desired and interfering signals. The baseband received signal and the parameter estimates are used in the likelihood information
generator 36 to calculate the likelihood functions corresponding to different QPSK symbol hypotheses corresponding to the desired and interfering signals. The likelihood information generator 36 could provide likelihood or log-likelihood functions depending upon the user's preference. The likelihood information processor 40 calculates the soft information values
corresponding to each bit. The partition of the functions described herein for the likelihood information generator 36 and the likelihood information processor 40 can be different
depending upon user preference. Also, these two blocks can be combined to obtain a single block which generates soft information using the channel parameters and the received baseband signal. Fig. 3 illustrates an alternative joint demodulator 32' where the baseband received signal and parameter estimates from the channel estimator 38 are used to obtain joint metric values. These are done using a joint metric computer 36' and a metric processor
40'. The metric processor 40' generates soft information values corresponding to the desired signal.
As described herein, the soft information generation can be applied to both non-dispersive channels, i. e.; flat-fading channels, and dispersive channels. For the π/4-
DQPSK modulation using IS-136, the user bit information is contained in the differentially
modulated symbols, but both the conventional equalizer and the joint demodulator use
coherent symbols in the MLSE/DFSE. In the conventional equalizer, the four hypothesized
states correspond to the second of two coherent symbols, i.e., the delayed coherent symbol. This results in sixteen different branch metrics that are calculated for each new received
sample. Calculating the optimal soft value for each detected bit requires estimation of the probability of that bit, which, in turn, requires the exponentiation and summation of the
metric values associated with that bit. As this is computationally expensive, sub-optimal
approaches using only the dominant terms from this calculation are employed. Limiting
operations, such as max() or min(), are used instead of the exp() operation. Even though these approaches are sub-optimal, they reliably estimate soft bit values for use in soft-input decoding.
Althoughjoint demodulation demodulates both desired and interfering signals together, soft information is calculated for only the desired signal. For optimal processing,
the interferers bit probabilities must be integrated out of the probability of the desired signals
soft bit estimate, requiring exponentiation and summation operations, that are to advantageously be avoided. Consequently, there are various different assumptions that can be made which lead to different sub-optimal approaches, described below.
For joint demodulation involving two users for a flat fading channel, the number .of possible metric values at each stage is 16 because QPSK has four possibilities for
each user and for two users the number of possibilities is 4*4=16. Therefore, the number of
possible metric pairs for the current and previous stage is 256; i.e., 16* 16=256. Among these 256 metric pairs, 128 will provide a +1 bit value for the desired differential bit, and the other 128 will provide a -1 bit value. One solution is to calculate all the metric pair sums for
the detected and nondetected differential bit values, then find the minimum values among them, and then find the difference of these minimum values to calculate soft information.
However, calculation of all the metric pair sums and finding the minimum among them is highly complex. For the nondetected bit, certain assumptions can be made in accordance
with the invention using various approaches.
A first such approach for a non-dispersive channel is illustrated in Fig. 5. The
detected bit and corresponding coherent symbol metric pair are found for the detected bit. In Fig. 5 the detected coherent symbols are D2 and 12 at stage n and D3 and 12 at stage n-1. (The symbols having the prefix D represent the desired bit, while the symbols having the prefix I represent the interfering bit. Each symbol value corresponds to a pair of differential
bits. For the nondetected bit, the coherent desired and interfering symbol values are kept as
is at the current stage n, i. e,, use the detected branch's metric for the current stage D2 and
12, and flip the coherent desired symbol value of the previous stage n-1 with the closest symbol that provides the nondetected bit value, while keeping the interferer as detected. This uses the symbol D4 and 12 in the example. Similarly, the coherent desired and interfering
symbols are kept as is at the previous stage, i. e., use the detected branch's metric for the
previous stage, D3 and 12, and flip the coherent desired symbol value of the current stage
with the closest symbol that provides a nondetected bit value, while keeping the interferer as
detected. This uses the symbol Dl and 12 in the example. Then, find the minimum metric
pair sums corresponding to these two symbol pairs as the dominant term for the nondetected bit.
For detected coherent desired symbol values, it may not be obvious how to find a closest nondetected value. Because of the interaction between the desired and
' interfering signals in the minimum metric calculation, there may be situations where the flipped desired symbol requires flipping the interfering symbol as well. It is possible to flip the desired coherent symbol value of the current or previous stage to the closest symbol value that provides the nondetected bit, while allowing the interferer coherent symbol value to float, i. e., take any value. For one interferer, this requires finding the minimum of 8
metric pair sums instead of a minimum of 2 metric pair sums.
This latter method can be extended further by also considering the other flipped (farther) coherent desired symbol and the minimum metric pair sum calculation for
the nondetected bit value. In other words, the closer and farther possible coherent sum
symbols are considered in flipping the detected coherent symbol. This assumes the minimum metric pair includes one of the minimum metric values of the current or previous states. In other words, the minimum of the current stage is fixed as is and it is necessary to find the minimum metric value from the previous stage that provides a nondetected bit value. In the
same way, the minimum of the previous stage is fixed as is and it is necessary to find the minimum metric value from the current stage that provides a nondetected bit value. The
minimum of these two possible pair sums is used as the dominant term for the nondetected bit. This approach requires finding the minimum of 16 metric pairs for the dominant term for
the nondetected bit.
Unlike flat-fading channels, where the metric pairs (branch metric pairs) are used for soft information generation, the path metrics (accumulated metrics) are used for soft information generation with dispersive channels.
In accordance with the invention the difference of the surviving path and two possible nondetected paths ("SOFT POS arid SOFT NEG") are calculated for each state.
I
The nondetected paths are obtained by flippirig the coherent detected symbols with the other border symbols while keeping the interferer Jsymbol as is. Once the SOFT POS and SOFT NEG values are obtained for each state the spft bit values are calculated using these values.
A first approach for dispersive channels is shown in Fig. 6, which extends the
I approach of Fig. 5 to dispersive channels. Fάr each state the difference of the surviving path and the two possible nondetected paths are Calculated (SOFT POS and SOFT NEG). The
nondetected paths are obtained by flipping jthe coherent detected symbol with the border symbols while keeping the interfering symbol as is. Assuming that the detected path is as shown in Fig. 6, and for a specific bit the nondetected paths are shown by the dashed lines,
I the soft information for the desired differential bit can be calculated similarly as
mm(abs(CM(n) -
+ l) - CMx(n + 1)))
Fig. 4 is a flow diagram of a routine implemented in the joint demodulator 32,
1 see Fig. 1, illustrating this approach. The apjproach begins at a block 50 which determines
I the jointly detected symbols and corresponding joint metrics. In equalizers, the detected path
I metric is found. At a block 52 a first potential nondetected metric sum is determined. This
I is done by flipping the current desired coherent detected symbol and keeping the interferer i coherent detected symbol as is. Both desired and interfering previous coherent symbols are
metric is found at the block 54. At a block 5J5 the difference of the detected joint metric and
the two nondetected joint metrics is found. Subsequently, the minimum of these two
I different metrics is found. In equalizers, the difference between the detected path metric and
I two nondetected path metrics is found and- then the minimum of these two different metrics
I is found. At a block 58 the soft values are generated in the conventional manner based upon
the minimum difference metric.
An extension of the approach described above relative to Fig. 6 narrows down • the approximation such that the desired flapped symbol allows change of the interferer
I
t s necessary to store 3 different values, soft X, soft Y and soft Z. The first two difference
values are the same as in the approach of Fig 7. The extra third difference value is obtained
from the farther flipped symbol.
Narrowing down the approximation even further provides a solution as
illustrated in Fig, 9. In the other approaches when the flipped path is calculated
corresponding to the stage n, the assumption was made that at stage n-1 both the detected and nondetected paths were merging. Assuming that the coherent symbol at stage n-1 is fixed, then to guarantee that, the flipped syirjbol at stage n comes from the fixed symbol at stage n-1, it is necessary to recalculate the ppth metric by forcing it to come from the fixed
symbol's path.
All of the described methods re an approximation to the dual-min (min/min) implementation of soft information generation,. Alternatively, approaches such as semi-MAP
I (Max a posteriori) based algorithms could improve performance over dual-min based approaches. In the semi-MAP based approach soft information is calculated per stage
instead of calculating soft information per state. If the soft values are not connected to any
specific state, there is no decision window aihd no path memory.
At the stage n+1, all the possible combinations of path metrics that provide a
detected bit value are calculated for all branches of all states. In the same way, all the possible combinations of path metrics that provide a nondetected bit value are calculated for
1 all branches. The minimum path metric corresponding to the detected and nondetected bit
values are obtained. Soft information is Calculated by finding the difference of these minimum path metrics. Since there is nc decision delay, the hard decisions degrade
perorm t e spec ed functions or steps, or combinations of specia purpose hardware computer instaictions.