PSP and AID Processing For The Next Generation Of Communications

1
PSP and AID Processing For The Next Generation
Of Communications

• JULY 2007

2
The Receiver Challenge
Transmitting a signal through a channel
introduces a myriad of potential degradations.
The degradations vary with environment and are
particularly difficult to mitigate if they are
rapidly time-varying.
3
Conventional Approach
Conventional receiver design is typically broken
into sequential functional modules - each
processing the results of the previous without
knowledge of the information used in the others.
4
Past Implementations
Conventional architecture employs separate
elements for front end processing (including
multiple antennas), demod and decoding. Each
element does a separate estimation resulting in
information being thrown away before the
decoding stage.
5
Past Implementations
Feedback in conventional systems represents a
valid attempt to share information between stages
but in rapidly time-varying scenarios feedback
information may no longer be valid, leading to
catastrophic failure.
6
TrellisWare Approach
Digital Likelihood Processing (DLP) combines all
stages in one fell swoop yielding better
results by using all of the information, all of
the time.
7
Optimal Channel Estimation
A truly optimum receiver would hypothesize every
possible combination of 1s and 0s running
equalization, synchronization/tracking,
demodulation, decoding, and any other channel
mitigation processing (e.g. interference
excision, or frequency estimation) on every
hypothesized data sequence.
8
Maximum Likelihood Process
The result would be a set of hypothesized receive
waveforms (based on each hypothesized sequence).
Each of these hypotheses would then be compared
to the actual received waveform and the closest,
or most likely, match chosen hence the phrase
likelihood processing
9
How use all sequences?
One obvious problem with this fully optimum
approach is that the number of sequences of 1s
and 0s that must be hypothesized grows
exponentially with the length of the sequence.
This is where TrellisWares variety of complexity
reduction techniques becomes critical simpler
implementation with minimal impact to the
ultimate performance.
10
Per Survivor Processing (PSP)
TrellisWares patented Per Survivor Processing
(PSP) prunes the ever expanding tree of possible
sequences to only the most likely sequences.
11
Trellis Derived By PSP
By eliminating the least likely sequence
extensions a trellis is created and the
processing per hypothesis described above can be
performed on each surviving sequence (hence per
survivor).
12
Conventional Digital Likelihood Approach
Conventional approaches use a similar natural
extension on Maximum Likelihood Sequence
Estimation algorithms (also referred to as the
Viterbi Algorithm). Data estimation is performed
on a trellis but uses the tentative decisions
from that processing to aid the channel
estimation and then feedback this information to
the Viterbi algorithm
13
Conventional Viterbi (Parameters estimated
globally)
While a step in the right direction, this
approach risks feedback delay problems in
dynamically changing channels since it creates
one global channel estimate that is then used for
processing all paths through the trellis.
14
PSP Trellis Likelihood Processing
PSP, on the other hand, updates channel estimate
and mitigation processing instantaneously for
each surviving sequence. Thus, not only is the
correct data used at the correct time, but the
history of each path can be used in the
estimation process for that path an essential
element to managing rapidly changing channels due
to things like high dynamics or interference. In
PSP, the correct path uses the correct data and
the correct history for each step in the trellis
estimation process.
15
Complex Modulations Waveforms Increase
Difficulty
PSP processing, while very effective, works best
with binary modulations. Brute force PSP can be
impractical for some modulations with high
spectral efficiency such as large QAM
constellations or designs using MIMO technology
for space-time multiplexing. This is because the
tree of possible sequences grows exponentially in
the base of the modulation cardinality.
16
Need New Approaches For Joint, Maximum Likelihood
Where the complexity of PSP is too great a new
set of techniques is required to approximate the
optimum joint maximum likelihood processing. One
approach is to approximate the signal with a
simpler signal where the deviations due to the
approximation are tracked and estimated using
PSP. For example, in the illustration here only a
subset of a QAM constellation might be used to
represent the signal in the PSP trellis, with the
deviation to other constellation points being one
of the parameters that is estimated in the PSP
trellis processing.
17
Large Interleavers Greatly Increase Complexity
For PSP to work effectively the tree must grow
sufficiently large before pruning to ensure that
the correct path is not pruned. If the tree
expands too rapidly the size of the PSP trellis
can still become intractable. For example if a
system has a large interleaver embedded in the
signal structure (e.g. for channel interleaving
or for turbo decoding) then the tree of possible
sequences grows exponentially with the size of
the interleaver.
18
Adaptive Iterative Detection (AID)
Another complexity reduction technique, called
AID, breaks the receive processing into pieces
much as in conventional receiver designs. Rather
than sequentially pass results from one piece to
the next, each piece makes preliminary, or soft
decisions which are shared with other Soft In
Soft Out (SISO) processing blocks of the
receiver. By correctly iterating between the
different pieces multiple times, convergence
towards the optimum joint solution is approached.

19
AID Was Motivated By Turbo Coding

• Turbocoding uses iteration to achieve outstanding
  performance at low SNR

Unlike conventional Forward Error Correction
(FEC) approaches, turbo-like codes are composed
of multiple simple constituent codes separated by
an interleaver. Decoding occurs by using
iteration to converge on the optimum result.
However, turbocoding deals only with FEC and is
not adaptive within a block of data.
20
Adaptive Iterative Detection (AID)
Dr. Chugg
Dr. Polydoros
Dr. Andreas Polydoros and Dr Keith Chugg, both
TrellisWare co-founders, pioneered the fields of
PSP and AID research. This work has been widely
reported in leading international journals and in
Dr. Chuggs book, Iterative Detection. I
addition to their continued technical leadership
roles at TrellisWare, both remain active academic
researchers in these and related fields Dr.
Polydoros as a Professor at the University of
Athens and Dr. Chugg as a Professor at the
University of Southern California (USC).
21
Iterative Detection
AID and PSP are not mutually exclusive. Often PSP
can be embedded in an AID receiver design to
adapt that piece of the design, resulting in
faster convergence and/or improved overall
performance.
22
AID Results (Fading Channel)
A single AID iteration in fast fading channels
results in a factor of 10 improvement. 4
additional iterations provide an additional 10 dB
of improvement and the typical turbo-like
performance curve where the Bit Error Rate (BER)
curve drops very rapidly.
23
AID and PSP the next generation of advanced
communication processing
Turbocoding radically changed the traditional FEC
in the 1990s by taking a significantly different
approach. AID and PSP are taking the next step -
enabling communications to connect in
environments previously considered too stressing.