Model Order Selection for LASSO Fitted Millimeter Wave Vehicular Channel Data

For analysis of vehicle-to-vehicle (V2V) communications, a firm understanding of the underlying channel is crucial. Through spatial filtering of narrow beams millimeter wave (mmWave) channels become sparse. As accurate channel models are required, without overfitting channel measurements, we investigate the optimal model order given V2V mmWave channel measurements. We use the complex Least Absolute Shrinkage and Selection Operator (LASSO) to find a sparse tapped-delay line representation of the measured channel, and apply the Akaike Information Criterion to find the optimal model order. Beyond this optimal order we start to fit measurement noise. Our results show, that for the given measurements, 4 to 8 taps prove to be an ideal model order.