Prediction of Fading Broadband Wireless Channels

1
Prediction of Fading Broadband Wireless Channels
JOINT BEATS/Wireless IP seminar, Loen

• Torbjörn Ekman
• UniK-University Graduate Center
• Oslo, Norway

2
Contents

• Motivation
• Noise Reduction
• Linear Prediction of Channels
• Delay Spacing, Sub-sampling
• Results
• Power Prediction
• Results
• Recommendations

3
Why?

• With channels known in advance the problem
  with fast fading can be turned into an advantage
• Adaptive resource allocation
• Fast link adaptation
• The multi-user diversity can be exploited

4
Noise Reduction of Estimated Channels
The estimated Doppler spectrum is low pass and
has a noise floor.
The same noise floor is seen in the power delay
profile.
5
IIR smoothers
6
FIR or IIR Wiener-smoother?

• IIR smoothers
• based on a low pass ARMA-model
• can be numerically sensitive
• need few parameters
• FIR smoothers
• based on a model for the covariance
• need many parameters
• Both have similar performance.
• Both use estimates of the variance of the
  estimation error and the Doppler frequency.

7
Linear Prediction of Mobile Radio Channels

• A step towards power prediction
• Can produce prediction of the frequency response

• Model for the tap
• The FIR-predictor
• The MSE-optimal coefficients

8
Linear prediction with noise reduction
9
Model Based Prediction
10
Delay Spacing
11
The MSE optimal delay spacing for the Jakes model
depends on the variance of the estimation error.
The NMSE has many local minima.
12
Sub-sampling and aliasing

• OSR 50
• Sub-sampling rate 13
• Jakes model

• SNR 10dB
• 16 predictor coefficients
• FIR Wiener smoother (128)

13
Prediction performance on a Jakes model

• OSR 50 (100 samples per l)
• FIR predictor, 8 coefficients

• FIR Wiener smoother (128)
• Dashed lines no smoother

14
The Measurements

• Channel sounder measurements in urban and
  suburban Stockholm
• Carrier frequency 1880MHz
• Baseband sampling rate 6.4MHz
• Channel update rate 9.1kHz
• Vehicle speeds 30-90km/h
• 1430 consecutive impulse responses at each
  location
• Data from 41 measurement locations

15
Prediction performance on the taps
16
Channel prediction performance
17
Power Prediction

• The power of a tap
• A biased quadratic predictor
• An unbiased quadratic predictor
• Rayleigh fading taps the optimal q for the
  complex tap prediction is optimal also for the
  power prediction.

18
Biased and unbiased NMSE
19
Observed power or complex regressors?

• AR2-process
• Approx. Jakes
• FIR predictor (2)
• Dash-dotted line for observed power in the
  regressors.

20
Power prediction performance
21
Median tap prediction performance
22
Channel prediction
23
(No Transcript)
24
Compare average predictor with unbiased predictor
25
Predictor Design

• Estimate the channel with uttermost care.
• Noise reduction using Wiener smoothers.
• Estimate sub-sampled AR-models or use a direct
  FIR-predictor.
• Estimate as few parameters as possible.
• Design Kalman predictor using a noise model that
  compensates for estimation errors
• Power prediction Squared magnitude of tap
  prediction with added bias compensation.