transmit beamforming for acoustic ofdm
TRANSCRIPT
Transmit Beamforming for Acoustic OFDM
Milica Stojanovic
Northeastern University
October 2020
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Hey, I’m talking to you!
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Transmit beamforming
I Why:- efficient use of power- avoidance of unintended listeners- spatial division of multiple users
I How: Assign a weight to each transmit element so that theirsignals add constructively at the receiver.
I Problem: Weights depend on the channel. Downlink channelmust be inferred from uplink. Estimate is noisy and delayed.
I What has been done: A lot in radio, a little in underwateracoustic systems (adaptive modulation, time reversal).
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
OFDM
I bandwidth scalability
I ease of FFT processing
I possibility of differentially coherent detection
I proven methods for acoustic Doppler compensation
I ideal platform for broadband beamforming
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Basic setup: Multiple transmitters
+
noise
RX
TX1
data in
TX2
TXM
.
.
.
data out
CHANNEL 1
CHANNEL 2
CHANNEL M
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Basic setup: Multiple paths
+
noise
RX
TX1
data in
TX2
TXM
.
.
.
data out
CHANNEL 1
CHANNEL 2
CHANNEL M
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Basic setup: Multiple frequencies
+
noise
RX
TX1
data in
TX2
TXM
.
.
.
data out
CHANNEL 1
CHANNEL 2
CHANNEL M
K frequencies
P paths
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Optimal beamforming (sorry, no pretty pictures)
I Hmk =
∑p h
mp γ
mp (fk)e−j2πfkτ
mp : channel transfer function on
the k-th carrier, m-th element
I wmk : beamformer weight on the k-th carrier, m-th element
I yk = dk∑
m wmk Hm
k + zk = dkwTk Hk + zk : signal received on
the k-th carrier
I Maximum SNR: wk ∼ H∗kI Normalization: 1
K
∑k w
Hk wk = 1
no extra power expenditure on account of beamforming
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Channel estimation via feedback (uplink)
noise
RX
TX1
estimatesout
TX2
TXM
.
.
.
pilots in
CHANNEL 1
CHANNEL 2
CHANNEL M
+
+
+
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Channel estimation via feedback (uplink)
noise
TX
RX1
estimatesout
RX2
RXM
.
.
.
pilots in
CHANNEL 1
CHANNEL 2
CHANNEL M
+
+
+
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Channel estimation
I Downink channel has to be estimated from the uplink pilot.
I If Hdn = Hup, the only problem is noise.
I During the time it takes to close the feedback loop (∼ 2d/c),the system geometry could change (ever so slightly), henceHdn 6= Hup.
I xmk = Hm,upk + zm,upk : received pilot signals
I Hdn = L(X) = Hup + αZup: estimated channel
I α = 0: noiseless estimate
α =√
LK , L = dBTmpe: LS estimation in the IR domain
α = 1: time reversal (TR)
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Beamforming with a channel estimate
I yk = dk∑
m wmk Hm,dn
k + zdnk = dkwTk H
dnk + zdnk
I Noises are characterized by σ2dn, σ2
up.
I W ∼ Hdn∗: Beamformer weights are still determinedaccording to the maximum SNR rule, but a channel estimateis used instead of the unknown true value.
I Case studies: perfect channel knowledge, delayed channel(noiseless estimate), TR, IR; no adjustment (wm
k = 1√M
).
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
How good will the channel estimate be?
I The time to close the feedback loop allows for the followingchanges to occur:(i) small-scale fading coefficients γmp(ii) system geometry (e.g. rx drifts at speed v in direction θR).
I Neither is fully predictable.
I Both can cause a significant change in the channel response.
I Grand question: Is there some feature of the channel thatchanges slowly enough that it can withstand the feedbackdelay, yet be exploited to formulate an efficient beamformer?
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Beamforming in the principal path’s direction
TX array
RXprincipal path𝜃0
d0
d0 sin 𝜃0
plane wave
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Beamforming in the principal path’s direction
I Principal path is stable (no surface interaction).
I Drifting will cause tx-rx positioning to change by a few metersover a few seconds, but this change is negligible compared toa transmission distance on the order of kilometers. Thechange in the principal path’s angle of arrival is thus expectedto be negligible.
I Note: receiver must still compensate for the Doppler shift(on either side of the link).
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Beamforming: Classical approach (array processing)
I Plane wave propagation, equally-spaced array elements (d0),signal coming from direction θ0:τm0 = τ0
0 + m∆τ0, m = 0, . . . ,M − 1∆τ0 = d0
c sin θ0
I If the signal is properly synchronized, τ00 = 0.
I For a narrowband signal of frequency f0, beamformer weights
are wm0 = 1√
Me j2πf0τ
m0 ∼ e j2πmf0
d0c
sin θ0 (or ej2πm
d0λ0
sin θ0)
I Acoustic communication signal is not narrowband, sobeamforming weights are set for each frequency fk ,
wmk = e j2πmfk
d0c
sin θ0
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Channel-based vs. angle-based beamforming
I Hmk =
∑p h
mp γ
mp (fk)e−j2πfkτ
mp =∑
p hmp γ
mp (fk)e−j2πfkτ
0p︸ ︷︷ ︸
hmp (fk )
e−j2πmfk∆τp
I wmk ∼ Hm∗
k =
hm∗0 (fk)︸ ︷︷ ︸h0e
−j2πfkτ00
e j2πmfkd0c
sin θ0︸ ︷︷ ︸steering to θ0
+∑
p 6=0 hm∗p (fk)︸ ︷︷ ︸
?
e j2πmfkd0c
sin θp︸ ︷︷ ︸steering to θp
I Optimal beamforming: match the phase and gain of everypath and array element.
I Beamforming in the principal direction: treat all other paths asnuisance (give up on predicting small-scale fading coefficientson the downlink), and just steer the beam in direction θ0.
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Estimating the principal angle of arrival
I Let wmk (θ) = 1√
Me j2πmfk
dc
sin θ
I Beamforming on the k-th carrier of the uplink signal wouldyield
∑m wm
k (θ)xmk = wTk (θ)xk.
I Total power after beamforming is S(θ) =∑
k |wTk (θ)xk|2
I Principal angle is estimated as that angle for which the poweris maximized: θ0 = arg max
θS(θ)
I Beamformer weight are set to wmk = wm
k (θ0).
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Notes on narrowband beamforming
I Narrowband beamforming rests on the assumption that thesignal bandwidth is much smaller than the center frequency.
I The weights assigned to different transmit elements are thesame for all signal frequencies.
I If the weights are evaluated for f0, then wmk = wm
0 .
I For angle-based narrowband beamforming,
θ0 = arg maxθ∑
k |wT0 (θ)xk|2, wm
k = wm0 = 1√
Me j2πmf0
dc
sin θ0
I Question: How much is the performance degraded undernarrowband assumption over underwater acoustic channels?
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Numerical illustration
I System geometry:d = 1 km, h = 100 m, hR = 20 m, hT = 70 m (top)v = 0.5 m/s, θR = 45◦ ⇒ d increases, hR decreases by0.47 m between uplink/downlink time
I Frequency occupancy: f0 = 10 kHz, B = 5 kHz
I Transmit array: M = 12, d0 = 0.345 m (2.3λmax or 3.45λmin)
I hmp , τmp calculated from system geometry; c=1500 m/s in
water, 1300 m/s in bottom
I Small-scale fading: independent across array elements;σs , σb ∼ λ0, Bδp = 10−4 Hz
I Signal: K = 1024, differential QPSK
I Noise: σ2up = σ2
dn = σ2, SNR= 1σ2 with 1
MK
∑m
∑k
|Hm,upk |2 = 1
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Multipath structure
delay [ms]0 0.005 0.01 0.015 0.02 0.025
uplin
k
-1
-0.5
0
0.5
1path gains
fist array elementlast array element
delay [ms]0 0.005 0.01 0.015 0.02 0.025
dow
nlin
k
-1
-0.5
0
0.5
1
fist array element
last array element
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Frequency response of the channel: Uplink vs. downlink
frequency [kHz]×10
4
1 1.05 1.1 1.15 1.2 1.25 1.3 1.35 1.4 1.45 1.5
chan
nel r
espo
nse
mag
nitu
de (
first
arr
ay e
lem
ent)
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
uplinkdownlink
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Frequency response of the channel: Across the array
frequency [kHz]×10
4
1 1.05 1.1 1.15 1.2 1.25 1.3 1.35 1.4 1.45 1.5
chan
nel r
espo
nse
mag
nitu
de (
uplin
k)
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
first arary elementsecond array element
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Performance results
SNR [dB]10 15 20 25
MS
E [
dB
]
-35
-30
-25
-20
-15
-10
-5
0
5
10
CB-BB, true channelCB-BB, delayed channelCB-BB, TR estimateCB-BB, IR estimateAB-BB, true angleAB-BB, estimated angleCB-NB, true channelCB-NB, delayed channelAB-NB, true angleAB-NB, estimated anglenone
CB/AB=channel based/angle based, BB/NB=broadband/narrowband
delayed channel = uplink channel, noiseless
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Inside angle-based beamforming: S(θ)
SNR=20 dBθ0 = 0.05 rad (0.05-0.0537 up; 0.0504-0.0542 down; 0.0524 est.)
θ [rad]-0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4
S(θ
)
×104
0
1
2
3
4
5
6
7
20000
40000
60000
80000
30
210
60
240
90
270
120
300
150
330
180 0
Re
-1.5 -1 -0.5 0 0.5 1 1.5
Im
-1.5
-1
-0.5
0
0.5
1
1.5
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Resolution and ambiguity limits
I λmaxM∆max
< d0 ≤ λmin∆min
∆max ,min = max, min(p,q){sin θp − sin θq}
I LHS: Paths p and q are resolvable.
I RHS: There is no ambiguity as to whether the angle of path pis x degrees above or below the angle of path q.
I Note: include only those paths that are stable and notnegligible in strength.
Milica Stojanovic Transmit Beamforming for Acoustic OFDM
Conclusion
I When the channel varies over the time it takes to close thefeedback, discrepancy between the uplink and the downlink issignificant enough to rule out channel-based beamforming.
I A possible solution is angle-based beamforming in thedirection of the principal path (that which has no surfaceinteraction and is stable). Rx/tx motion over theuplink-downlink should not cause a significant change in theangle of a typical geometry.
I Next steps:angle trackingbroadband null steeringmultiple userssimultaneous tx/rx beamforming.
Milica Stojanovic Transmit Beamforming for Acoustic OFDM