Soft Channel Estimation

Ballard BlairMIT/WHOI Joint Program

January 3, 2007

1/3/2008 1

Typical Scattering Function

1/3/2008 2

Depth = 15mDistance = 225m

Preisig, SPACE02

Underwater Signal Paths

1/3/2008 3

Dynamic Channel

1/3/2008 4

Wave Height

Signal Estimation Error (SER)

Impulse Response

Turbo Equalization

1/3/2008 5

MAPEqualizer Interleaver

MAPDecoder

Deinterleaver

Data

Channel Model

• Finite impulse response (FIR) channel

Typical channel length, M≈50 for 12kHz carrier, 100m depth, 4kbps data rate

1/3/2008 6

Soft Channel Model (Song2004)

• Data Model:

• Cost Function:

• Solution:

1/3/2008 7

Soft RLS Algorithm

1/3/2008 8

Matching Pursuit

1/3/2008 9

Other variations: Orthogonal MP (OMP), Order Recursive Least Squares MP, etc.

Problems

• RLS does not take advantage of sparsity

• Matching pursuit not adapted for soft data

• Errors in matching pursuit “dictionary” still an open problem

1/3/2008 10

Ideas

• Use two step approach– Identify energetic taps (maybe matching pursuit)– Use RLS technique with only those taps

• Use weighted matching pursuit– Directly adapt matching pursuit to handle soft

data

1/3/2008 11

More ideas?

• Compressed sensing?– Does not use all of our knowledge about channel,

but still might be good?

• Something else?– Need technique that uses soft data and sparse

channel model

1/3/2008 12

Questions?

1/3/2008 13