1. Design and Development of a Software-Defined Underwater Acoustic Modem for Sensor Networks for Environmental and Ecological Research 
Tricia Fu, Daniel Doonan, Chris Utley, Ronald Iltis, Ryan Kastner, and Hua Lee

2. Outline Summary of MP Algorithm and Aquamodem Implementation
Comparison w/ other Acoustic Modems
Hardware Implementation
Test Results: wire, air, tank
Future ideas (in the works)

3. Design Objectives Operating range: >= 100 m
Operating depth: up to 20 m
90-day deployed lifetime
Handle multipath spread of several ms (up to 11.2 ms)
What else to add here?

4. MP Algorithm Changes from last paper/presentation:
Inclusion of channel order estimation (NF estimation).
Improved timing acquisition by increasing number of training symbols used for synch: in most test cases, from 2 to 5.

5. MP Algorithm: Simulation
Simulation results, with additions to prior MP-algorithm made.
Estimated Multipath Intensity Profile (MIP) from simulations, actual number of multipaths = 10, SNR = –4 dB. Left: 2 training symbols, and
estimated Nf varying from 1 to 16 with SER = Right: 5 training symbols, and estimated Nf varying from 5 to 16 with SER =
Estimated MIP from simulations, with actual number of multipaths = 10 and SNR = -4 dB.
Left: 2 training symbols, SER =
Right: 5 training symbols, SER =

6. Matching Pursuits vs FSK
Direct Sequence Spread Spectrum (DSSS) preferred over frequency hopping when channel exhibits deep, narrow fades so that signal energy spread over wide bandwidth.
MP produces channel estimates during symbol demodulation process.
Add – check Iltis paper.

7. Signaling Scheme 8-ary code orthogonal modulation
56-chip Gold sequences, 6000 chips/sec
Chip sequences applied to root-raised-cosine filter, 30 % excess bandwidth
24 kHz center frequency
Guard interval, normally same length as symbol duration.
At Fs = 96 kHz rate, symbol rate is ~ 54 Hz, bit rate ~ 161 bps.

8. Hardware Implementation
Texas Instruments’ C6713 DSP Board
Sonatech custom built transducers: horizontal omnidirectional coverage in 20 kHz – 30 kHz range
System power: alkaline primary cells
Commercial single-chip class D power amp driving matching network, automatic transmit/receive switch.
Tx transducer also connected to low-noise variable gain amplifier when power amp not operating.

9. Hardware Implementation
Blah need djd input!

10. Wire Test: results From 2006-07-22-wiretest folder,
Test Setup: Wire test, audio cable between 2 C6713 boards, 96 kHz Fs, 100 packets, 43 pack/sym, hypothesized NF ranged from 1 to 5 w/ only 1 “real” path, RVSync symbol was 5 * normal symbol length, so total RVSync vector (w/ guardband) was 6 NS
Error rates: Always 0 symbol error rate
Things to note: very clean MIP; Note that synch done to align start of symbol at around 5 samples delay, or ?? ms. Not understood is the harmonics in doppler spread spectrum plot

11. Air Test in Lab: results
From L:\Research\ airtest_2\ airtest, t3
* Note DSP sampling rate decreased from 96 kHz to 8 kHz for use with off-the-shelf audio components (speaker, microphone)
Generally zero error rate, again tests held as proof of concept in the relatively low SNR environment, but with real multipaths introduced.
Doppler spread w/ 4-6ft distance between microphone, speaker was ~ Hz

12. Air Test in Hallway: setup
Pictured: Left: Hallway with Rx setup in front, and Tx setup in the rear (highlighted in red). Right: closeup of Tx setup, board and computer not shown.
2nd bullet: L:\Research\ hallwaytests8khz; even then, low error rate, nonzero error rate occurred once out of series of 4 consecutive tests held on 1 day.
3rd bullet: L:\Research\ airtest_2\ airtest; errors w/ data gathering so
67.5 ft between Tx speaker, Rx Microphone, 54 ft between Tx speaker and closest “end” wall
For initial tests w/ fixed NF, synch w/ 2 symbols, 0 error rate unless doors were opened or people walked by mid-test.
For later tests w/ improved synch and NF estimation, see table to the right; 117 ft between Tx spkr, Rx mic
Earlier Tests
Later Tests
Range
67.5 ft
117 ft NF
Fixed
Estimated
# training symbols 2 5

13. Air Test in Hallway : results
MIP shown here: test estimated NF ranged from 4-11 (117 ft between Rx, Tx), max NF set to 30
L:\Research\ hallway_with_corrected_nf_estimation, test # 6 (.xl sheet tracking data/figs used said test #7, but should be #6)
Trial
SER
Doppler Spread (Hz) 1
0.0233
0.173 2
0.1903 3
0.1557

14. Air Test in Stairway: setup
The test setup placed the receiver on the third floor, with the transmitter on the fourth floor. The stairwell connected the first through fifth floors. Test duration was again 6 packets, 43 symbols per packet, with sampling rate set to 8 kHz since standard audio equipment was used.
Clockwise, from top left: Receiver setup on 3rd floor, view from middle of 4th and 3rd floor stairway, receiver setup on 4th floor.

15. Air Test in Stairway: results
L:\Research\ stairway, test #2 pictures,
For each test, the table above lists the guard time as a multiple of the transmit duration, the symbol error rate, and the Doppler spread. At the time of the test, no Nf estimation had been implemented, so Nf had been fixed to a high value.
Low Doppler spread low due to the much lower symbol rate resulting from the longer guard time.
MIP’s horizontal axis reduced from 1200 ms to 500 ms due to lack of paths in the delay range greater than 500 ms.
Trial NF
Guard time
SER
Doppler Spread (Hz) 1 20 9
0.0078
0.0069 2 11
0.0029 3 30

16. Tank Test at MSI: setup Need better pic from Chris (get Friday)
Tank tests: UCSB’s Marine Biotechnology Laboratory, seawater processing room,
* Fs = 96 kHz, 24 kHz center freq used w/ transducers, 6 kHz 2-sided bandwidth
* 2 series of tests: Series A, before improvements to sync, also no NF estimation, so fixed NF, 2 training symbols for sync. Also, 0.6 and 12.7 m distances.
Tank access through hatches precluded the use of intermediate distances.
Series B discussed later.
Need better pic from Chris (get Friday)

17. Tank Test at MSI: noisy environment
Step frequency test: from this we noted noise was not white nor Gaussian
0.60 m distance between transducers
Tx sending step chirp, 20 kHz:0.50 kHz:28 kHz, 2 ms per frequency
Rx captured 34 ms unprocessed ADC samples (3264 A/D samples at 96 kHz sampling), dumped contents from mem to hex file, then used Matlab to generate picture above.
Estimated SNR about 6.1 dB at 0.60 m distance; no SNR estimate at 12.7 m distance, but must’ve been considerably lower.
Hardware problems w/ second series (B) of tests, so SNR during B series tests was -1 dB.

18. Tank Test Series A: video
To get from Chris (20-30 sec) on Friday.

19. Tank Test Series A: error rates, results
Left: 0.60 m distance MIP, 6.1 dB SNR – note the weaker path at 60 samples, or about 6 ms.
Right: 12.7 m distance MIP, unmeasured SNR, but obviously lower
Note: I probably should change Delay from samples to ms!
Table of SER, Doppler Spread: note 12.7 m higher SER due to synch errors, which show up as offset chunks in the MIP. Also, NF = 10 clearly is overestimating # of significant multipaths at 12.7 m distance.

20. Tank Test Series A: Doppler Spread
Note: maybe should normalize these plots somehow?
* ~ 0.2 Hz Doppler Spread for 0.6 m distance (we consider this to be reasonably accurate)
* ~ 0.9 Hz Doppler spread (likely incorrect due to synch errors) for 12.7 m distance
Doppler spread, t5 (0.60 m), and t10 (12.7 m)
Doppler Spectrum Plots, at 0.6 m distance and 12.7 m distance
Doppler Spread: 0.2 Hz vs 0.9 Hz

21. Tank Test Series B 5 training symbols per sync, up from 2
NF estimation code integrated
Lower SNR due to Tx power amp problems (-1.0 dB vs. 6.1 dB)
Only 0.60 m results collected
* Series B tests: increased # training symbols (5), also NF estimation, but hardware issues w/ Tx power amp resulted in very low SNR and thus 12.7 m tests had no results.
Left: DSP boards, transmitter power amplifier and battery.
Right: Top view of one of the transducers, partially submerged in saltwater tank.

22. Tank Test Series B: results
Tank Test A (t5, L:\Research\ pooltest), Tank Test B (t6, L:\Research\ tanktest2)
Previous MIP at 0.60 m distance pictured here with Series B MIP at 0.60 m distance.
Note weak path at 6 ms not as apparent.
Series A: 6.1 dB vs. Series B: -1.0 dB SNR
Note weak path at 6 ms not as apparent in Series B MIP compared with Series A MIP.

23. Doppler Spread Plots Comments to make?
Wire test plot, hallway air test, Series A tank test, Series B tank test normalized Doppler Spread plots

24. Acknowledgments
MSI, Keck Foundation, Navy, Sonatech, blah