1. 2010 IEEE Workshop on Signal Processing Systems
Multi-Stage and Sparse Equalizer Design for Communications Systems in Reverberant Underwater Channels
Karl Nieman, Kenneth Perrine, Terry Henderson, Keith Lent, Terry Brudner, and Brian Evans
Applied Research Laboratories: The University of Texas at Austin
2010 IEEE Workshop on Signal Processing Systems
Oct. 8, 2010

2. Receiver Communicator(s)
Buoys
Receiver
Divers
Communicator(s)
Seafloor Instruments
UUVs

3. Reverberations are significant
-20
-10 10 20 30 40
-70
-60
-50
-40
-30
Representative Underwater Reverberation
Sample time (ms)
Signal strength (dB)

5. Doppler effects can be horrendous
Toward
Away
???

6. Phase with no Doppler effects50
100
150
200
250
-pi
-pi/2
pi/2 pi
QPSK Ideal
Sample time (ms)
Phase 00 01 10 11

7. Uncompensated Doppler effects50
100
150
200
250
-pi
-pi/2
pi/2 pi
Sample time (ms)
Phase

8. After gross Doppler compensation!! 50
100
150
200
250
-pi
-pi/2
pi/2 pi
Sample time (ms)
Phase 00 01 10 11

9. Doppler Detection: Carrier Recovery
f(t) t t
g(t) = f(t)2
Expected for
Zero-Doppler
Doppler-inflicted
Observation
FFT ω
|G(ω)|
O(M ln M) for M samples

10. Doppler Detection: Replica Bank… …
O(cM ln M) for
M samples and
c replicas
Δf = 3.6 Hz
c replicas
Δf = 3.4 Hz
Δf = 3.2 Hz … …

11. Doppler Detection: Marker Distances
O(M ln M) for M samples

12. Static Equalizer
Decision
Feed-forward taps Σ
x[t]
y[t]
Feedback taps

13. Sparse Equalizer
Decision
Feed-forward taps Σ
x[t]
y[t]
Feedback taps

14. Fully Adaptive Equalizer
Decision
Feed-forward taps Σ
x[t]
y[t] –
Update
Feedback taps
Update: O(N) per symbol
(N = total # of taps)

15. Single-Tap Adaptive Equalizer
Decision
Feed-forward taps Σ
x[t]
y[t] –
Update
Feedback taps
Update: O(1) per symbol

16. Which is the best combination to use?

17. July, 2009
Start
250 m
150 m
- BPSK, 30 kHz carrier, 1 kHz symbol rate (1 kbit/sec)
- QPSK, 25 kHz carrier, 32 kHz symbol rate (64 kbits/sec)

18. Results: Gross Doppler Detection
Carrier recovery  highest accuracy

19. Results: BPSK (Narrowband) Packet
SNR at 150 m, Stationary
SNR at 250 m, In Motion
# feedback taps
# feedback taps
Static EQ
Sparse EQ
Adaptive (Single-Tap)
Adaptive (Full)

20. Results: QPSK Packet SNR at 150 m, Stationary SNR at 250 m, In Motion
# feedback taps
# feedback taps
Static EQ
Sparse EQ
Adaptive (Single-Tap)
Adaptive (Full)

21. Adaptation rate (8 feedforward and 32 feedback taps)
BPSK, stationary
BPSK, in motion
QPSK, stationary
QPSK, in motion
Output SNR (dB)
Adaptation rate (8 feedforward and 32 feedback taps)

22. Conclusions No single choice for EQ parameters
EQ type
Number of adaptive taps
Adaptation rate
Underwater channel changes quickly
Reverberation length
150m to 250m
Underwater features
Thermocline

23. Conclusions Fully adaptive equalizer:
Most sophisticated and most computationally expensive
Does not work in all cases
Run multiple techniques and choose the best

24. Dataset 5-element underwater dataset available:
5 simultaneous receiver elements
62.5 kHz center frequency, up to kHz bandwidth
MATLAB format
underwater/datasets/index.html