GSM Enhanced Full Rate Speech Codec

Description

The GSM Enhanced Full Rate (EFR) codec is a speech compression algorithm standardized for the GSM system. Its technical specification is contained in 3GPP TS 26.093. The codec operates at a bitrate of 12.2 kbit/s, which is the same as the original GSM Full Rate (FR) channel, but it utilizes a far more advanced coding scheme based on Algebraic Code Excited Linear Prediction (ACELP). The encoding process works on 20 ms frames of speech sampled at 8 kHz. For each frame, the encoder performs linear predictive analysis to extract parameters representing the vocal tract filter (10 LPC coefficients). It then searches an algebraic codebook (a fixed codebook) and an adaptive codebook (representing pitch) to find the best excitation signal that, when passed through the LPC filter, reproduces the original speech with minimal error. The parameters for the filter and the indices for the best codebook entries are quantized and transmitted as a 244-bit block every 20 ms, resulting in the 12.2 kbit/s rate. At the decoder side, these parameters are used to reconstruct the excitation signal and synthesize the speech through the LPC filter. Post-processing filters are applied to enhance the perceptual quality of the output. The EFR codec is designed to be robust to transmission errors. It employs unequal error protection, where the most perceptually sensitive bits in the encoded frame are given the strongest channel coding (convolutional coding), while less critical bits receive weaker protection or none. This ensures graceful degradation of voice quality under poor radio conditions. The codec is implemented in the mobile station's digital signal processor (DSP) and in the network's transcoding unit, often located in the Mobile Switching Center (MSC). When two EFR-capable mobiles communicate, the speech is encoded and decoded only at the endpoints, preserving quality. If communicating with a non-EFR device, the network performs transcoding to a compatible format.

Purpose & Motivation

The GSM-EFR codec was developed to address the primary criticism of early digital cellular voice quality, which was often described as 'robotic' or 'synthetic' compared to analog systems or landlines. The original GSM Full Rate (FR) codec, while a marvel of its time, used a relatively simple Regular Pulse Excitation-Long Term Prediction (RPE-LTP) scheme that had clear quality limitations, especially in the presence of background noise or for certain speaker types. The motivation for EFR was to elevate the voice quality of GSM networks to a level comparable to wireline telephony, thereby improving customer satisfaction and making GSM more competitive. It solved the problem of poor perceived quality without requiring any changes to the underlying GSM radio channel structure or bitrate, as it fit into the existing 13 kbit/s traffic channel (22.8 kbit/s after channel coding). This was a crucial design constraint, allowing for a software-upgradable deployment in networks and handsets. The development of EFR was driven by competition and the desire for a 'premium' voice service. It addressed the limitations of the FR codec by employing state-of-the-art ACELP technology, which provided superior modeling of the speech signal and excitation. Its introduction in Rel-8 (as part of the consolidated 3GPP specifications, though originally developed earlier by ETSI) marked a significant milestone in making high-quality digital mobile voice a reality, paving the way for later codecs like AMR which built upon its principles.