Implementing Linear Predictive Coding based on a statistical model for LTE fronthaul

Abstract: This thesis studies the application of Linear Predictive coding (LPC) in the downlink of Long Term Evolution (LTE) fronthaul, which comprises of BBU and RRH. This can act as an additional module in the existing system. Today, the transmission of a single complex sample from the BBU to the RRH consumes 30 bits. The research of the thesis is to analyze the application of linear prediction theory in the LTE downlink transmission, which will work as a compression scheme and reduce this 30 bits to lower value, at the same time fulfill the Error Vector Magnitude (EVM) requirement stated in the LTE standards made by 3rd Generation Partnership Project (3GPP). As 4G-LTE and the upcoming access technologies will deal with large number of data samples in the transmission, it is an advantage if those data samples can be compressed without destroying the information content. LPC or linear prediction coding has been proved to be a very effective method for speech compression in audio related applications. In this thesis, the same logic of compression is applied on digital data samples of the LTE and the results are analyzed. It is found that, if LPC is applied properly on the LTE, it is possible to compress data samples efficiently and transmit them from the BBU to the RRH with fewer bits. At the RRH those compressed data samples can be processed and the main information data can be reconstructed, with additional quantization error and noise. This is obvious because LPC is a lossy compression method. A statistical model is established to generate a table of linear prediction filter coefficients which will be present both at the BBU and the RRH, when compression and decompression of data samples are performed. Entropy is also calculated in order to analyze the achievable compression on an actual error vector after implementing certain compression coding such as Huffman coding. The specific coding technique is left as a scope of future research.