applied research laboratories the university of texas at austin 1 doppler estimation and correction...

APPLIED RESEARCH LABORATORIES

THE UNIVERSITY OF TEXAS AT AUSTIN

1

Doppler Estimation and Correction for Shallow Underwater Acoustic

Communications

Kenneth A. Perrine*, Karl F. Nieman*, Terry Henderson*, Keith Lent*, Terry J. Brudner*, and Brian L. Evans†

*Applied Research Laboratories: The University of Texas at Austin†Dept. of Electrical & Computer Eng., University of Texas at Austin

Asilomar Conference on Signals, Systems, and ComputersNov. 9, 2010



2

Users

Access Point

UUVsSeafloor

Datalink

Buoys

Divers

Underwater Acoustic Network



3

Underwater Acoustic Channel

• Propagation speed 200,000x slower vs. RF in air• Lowpass (bandwidth decreases with range)• Wideband communication relative to carrier• Shallow water case

Time-varying Doppler

Channel reverberationHigh energy

Long time constant

Measured shallow water channel impulse responsesRange is 30m for position 1 and 1260m for position 3.



4

Underwater Acoustic Channel

• Doppler effects for received QPSK signalResults from linear bulk Doppler correction

Decision regions



5

Proposed Contributions

• Shallow underwater acoustic communicationsOne-element transmitter (stationary and moving cases)

Quadrature phase shift keying (QPSK)

Carrier frequency 62.5 kHz and 31.25 kHz bandwidth

Transmit 31.25 kbps at distances of 30 to 1285 m

One-element receiver (anchored on floating platform)

1. Evaluate SNR performance of three Doppler estimation methods

2. Evaluate static and adaptive equalizers



6

Bulk Doppler Estimation

• Approach 1: Self-referenced correlationTransmit two copies of training sequence

Use phase in cross-correlation of received symbols

Rep. 1 Rep. 2

Payload…

Calculate phase offset in decoded symbols

Symbols

P. Moose, “A technique for orthogonal frequency division multiplexing frequency offset correction,”IEEE Transactions on Communications, vol. 42, no. 10, pp. 2908-2914, Oct. 1994



7


• Approach 2: Carrier recoveryObserve peak FFT

frequency of squaredsamples (in binaryphase shift keying(BPSK) case)

Compare observedfrequency withexpected centerfrequency (withoutDoppler)

t

f(t)

t

g(t) = f(t)2

t

g(t) = f(t)2

t

g(t) = f(t)2

Expected for

Zero-DopplerDoppler-inflicted

Observation

FFT

ω

|G(ω)|

Expected for


ObservationExpected for


Observation

FFT

ω

|G(ω)|



8


• Approach 2: Carrier recoveryVariation: slice packet into “windows”

Rough adaptation to time-varying Doppler effects

t

f(t)

t

g(t) = f(t)2

Expected for




Observation

FFT

ω

|G(ω)|

Expected for




Observation

FFT

ω

|G(ω)|



9


• Approach 3: Pilot toneEncode pure tone outside of data band

Average over all measured pilot frequencies to estimate deviation from transmitted frequencies

87 kHz tone+/- Doppler

Data:62.5 kHz center;

31.25 kHz BW



10


• Approach 3: Pilot toneVariation: slice packet into “windows”:

87 kHz tone+/- Doppler

Data:62.5 kHz center;

31.25 kHz BW



11

Windowing Tradeoffs

• QPSK decoding

250 ms 125 ms each 62.5 ms each 31.25 ms each



12

Packet Structure

• Linear frequency modulated (LFM) chirpResistant to Doppler

• Training – 128 symbols4 length-13 Barker sequences

76 symbols for equalizer training

Symbol rate of 15.625 kHz

• Payload – 3968 symbols• Guard interval at end

100 ms for reverberation analysis

• Pilot tones at 45 and 87 kHz

Packet Structure

Packet Spectrum



13

Experimental Setup

• Applied Research Laboratories Lake Travis Test FacilityLake

37 m depth

Former riverbed

Nearby dam

Transmitter on research vessel

Receiver on barge at test station



14

Data Collection Points

1: 15mdocked

2: 325-375mfloating

3: 1235-1285mfloating

4: 185-255mvertical motion

5: 300-80mtowing at ~3 kts



15

Static Equalizer

ΣFeedforward taps

Feedback taps

x[m] y[m]

Decision

5 feedforward taps3 feedback taps



16

Fully Adaptive Equalizer

ΣFeedforward taps

Feedback taps

x[m] y[m]

Decision

5 feedforward taps3 feedback taps0.01 learning rate

Update

–

Update: O(N) per symbol(N = total # of taps)



17

Issues with Windowing• Support for Doppler estimation accuracy decreased• Smaller samples are subject to more noise• Discontinuities (even when smoothed) can lead the

adaptive decision feedback equalizer (DFE) astray• Windowing mostly benefits static equalizer

Successful operation

Problematic



18

Experimental Results

• Average estimated SNR for bulk Doppler detection/correction and equalizationCarrier recovery (BCDE) provides highest SNR.

Adaptive equalizer has best increase in SNR overall

A: Self-referenced correlation

B, C, D, E: Carrier recovery(1, 2, 4, 8 windows)

F, G, H, I: Pilot tone(1, 2, 4, 8 windows) Bulk Doppler Detection Method



19

Experimental Results• Self-referenced

correlation (A) performs poorlyRepresents tiny

packet sample• Pilot tone

tracking (FGHI) performs poorly in motion case (Pos. 2)

• Carrier recovery with any number of windows (BCDE) performs best



20

Conclusions

• Windowing for Doppler detection benefits static equalization

• Pilot tone method was not reliable

• Best configuration over entire datasetSingle window carrier

recovery method

Adaptive equalization

t

f(t)

t

g(t) = f(t)2

t

g(t) = f(t)2

t

g(t) = f(t)2

Expected for


Observation

FFT

ω

|G(ω)|

Expected for




Observation

FFT

ω

|G(ω)|

Σ

Update

–



21

Underwater Acoustic Comm. Dataset

• Experimental Setup1-element transmitter

BPSK, QPSK, 4-QAM, 16-QAM and 256-QAM

Symbol rates of 3.9 and 15.6 kHz

With and without pilot tones

Ranges 10m to 1285 m

5-element receiver array in L shape

• Raw data in MATLAB formathttp://users.ece.utexas.edu/~bevans/projects/

underwater/datasets/index.html



22



23

Publications and PresentationsConferece Proceedings• K. F. Nieman, K. A. Perrine, T. L. Henderson, K. H. Lent, T. J. Brudner and

B. L. Evans, “Wideband Monopulse Spatial Filtering for Large Array Receivers for Reverberant Underwater Communication Channels”, Proc. IEEE OCEANS, Sep. 20-23, 2010 Seattle, WA

• K. F. Nieman, K. A. Perrine, K. H. Lent, T. L. Henderson, T. J. Brudner and B. L. Evans, “Multi-stage And Sparse Equalizer Design For Communication Systems In Reverberant Underwater Channels”, Proc. IEEE Int. Workshop on Signal Processing Systems, Oct. 6-8, 2010, Cupertino, CA

• K. A. Perrine, K. F. Nieman, T. L. Henderson, K. H. Lent, T. J. Brudner and B. L. Evans, “Doppler Estimation and Correction for Shallow Underwater Acoustic Communications”, Proc. Asilomar Conf. on Signals, Systems, and Computers, Nov. 7-10, 2010, Pacific Grove, CA

Released Dataset• “The University of Texas at Austin Applied Research Laboratories Nov.

2009 Five-Element Acoustic Underwater Dataset”, Version 1.0, 6-4-20105-element samples of BPSK, QPSK, 16QAM, 64QAM, and 256QAM signalsUp to 1300 yard range, up to 63 kbit/sec data rate



24

Data Collection

• TransmitterOmnidirectional transducer

Submerged between 1-8m

• Receiver4.6m depth

Five directional hydrophones

Half-power beamwidths• Horizontal: ~45°• Vertical: ~10°

• Sampling rate: 500 kHz

Transmitting TransducerSensitivity at 1m



25

Software Receiver

• Frame synchronizerIdentify LFM chirps via cross-correlation

• Bulk Doppler detection• Bulk Doppler correction

Linear interpolation of oversampled basebanded signal

• Decision feedback equalizer (DFE)Static

Decision-directed adaptive w/ learning rate of 0.01



26

July Raytracing

• Severe thermocline:Receiver R can’t directly see

transmitters A or B

Surface

Lakebed



27

Channel Impulse Response

Fig. 4. Channel impulse responses (CIR) for near and far ranges. Position1 range is 30 m and Position 3 range is ~1260 m.



28

Experimental ResultsA: Self-referenced correlation


F, G, H, I: Pilot tone(1, 2, 4, 8 windows)

Pos. 1: 15m, docked

Pos. 2: 325-375m,free floating



29

Experimental ResultsA: Self-referenced correlation


F, G, H, I: Pilot tone(1, 2, 4, 8 windows)

Pos. 4: 185-255m,vertical motion

Pos. 5: 300-80m,towing at ~3kts



30


• A BER of ~0.5 indicates catastrophic failure in decoding.

• 4 or 8 windows significantly helps the static EQ;

• However, adaptive EQ yields better results overall.



31


• Pilot tone approach was not be reliableMultipath interference caused selective fading

Pilot tone was too narrow in bandwidth

applied research laboratories the university of texas at austin 1 doppler estimation and correction...

Documents