Signal enhancement of radio frequency power measurement in 1/f noise

Abstract

One of the prevalence challenges in the radio frequency (RF) power sensor development is to reduce noise in the acquired signal. The noise in the signal subsequently contributes to the error in the measurement of signal parameters. Sources of noise could come from the chain of signal conditioning and acquisition in the sensor circuitry. The assumption of additive white Gaussian noise (AWGN) to model a measurement is not valid since many applications such as in RF power measurement has the noise coloured with 1/f spectrum characteristics. With this characteristics, the assumption of independent and identically distributed (IID) used in signal detection and estimation becomes not valid. By whitening process, the 1/f noise characteristics can be converted to be similar to white noise. The analysis results of decimation, linear prediction, Burg algorithm and chopper with averaging shows that the proposed methods can be used effectively. Both Burg algorithm and linear prediction are more complex due to the need to perform matrix inversion. The decimation and chopper with averaging are the least complex but it could only meet the requirements if the sample size is more than 300 samples. After performing the whitening, the wavelet transform and de-noising are implemented to remove noise as much as possible while preserving the signal characteristics. As results, it can be seen that the noise is removed while the characteristics of the pulse signal is preserved by the Haar wavelet. However, the recovered signal is distorted when using Daubechies 5 wavelet with significant reduction in noise. Based on the result for RF power measurement for different whitening methods in Monte Carlo simulation, Burg algorithm yields the highest total variance reduction which is 97.13%, followed by the linear prediction which is 90.3%, decimation 64.11% and lastly, chopper with averaging, 3.10% for SNR of 8 dB. Although Burg algorithm is more complex compared to decimation, it preserves all the signal samples which is more suitable for pulse signals and it is the best whitening method used in this research.