Enhanced Full Rate speech codec

Description

The GSM Enhanced Full Rate (EFR) speech codec is a 12.2 kbps codec standardized by 3GPP and ETSI. It employs the Algebraic Code Excited Linear Prediction (ACELP) coding technique, which was a state-of-the-art method at its introduction. The codec processes 20 ms speech frames, each containing 160 samples at an 8 kHz sampling rate. For each frame, the encoder analyzes the speech signal to extract parameters representing the linear predictive filter (LPC), the adaptive codebook (pitch), and the fixed algebraic codebook (innovation). These parameters are then quantized and transmitted over the radio interface at a net bitrate of 12.2 kbps, which fits into the GSM channel's gross bitrate after channel coding for error protection.

The EFR decoder on the receiving side reconstructs the speech signal by passing the excitation signal (derived from the adaptive and fixed codebooks) through the synthesis filter (defined by the LPC parameters). A critical component of EFR is its robust channel coding and error concealment mechanisms. It uses a Rate 1/2 convolutional code for forward error correction (FEC) and includes sophisticated algorithms to detect and mask frame errors, ensuring acceptable voice quality even in moderate radio channel conditions. The codec is designed to be interoperable and is mandatory for GSM terminals supporting the relevant frequency bands.

EFR's architecture is integrated into the GSM Full Rate traffic channel (TCH/FS). When a call is set up, the network and mobile station can negotiate the use of EFR if both support it, otherwise falling back to the original Full Rate (FR) or Half Rate (HR) codecs. The codec's performance is characterized by high Mean Opinion Score (MOS) ratings, often exceeding 4.0 under clean channel conditions, which was a substantial improvement over the FR codec's MOS of around 3.5. This made EFR a key enabler for high-quality commercial mobile voice service.

Purpose & Motivation

The EFR codec was developed to address the primary user complaint in early GSM networks: voice quality that was noticeably inferior to fixed-line telephony. The original GSM Full Rate (FR) codec, while a revolutionary digital codec for its time, had limitations in audio fidelity, particularly in the presence of background noise or for certain speaker types. The competitive landscape and user expectations demanded a mobile voice service comparable to wireline quality.

Its creation was motivated by advancements in speech coding algorithms, notably the ACELP technique, which offered a better trade-off between bitrate, complexity, and quality. The goal was to develop a codec that could use the existing GSM channel structure (the Full Rate channel) without requiring changes to the radio infrastructure, ensuring a smooth upgrade path. EFR provided this by fitting its 12.2 kbps payload into the channel capacity after applying similar channel coding overhead as the FR codec.

EFR solved the problem of perceived voice quality, which was critical for customer satisfaction and network differentiation. It allowed operators to market 'landline-like' voice quality. Furthermore, its robustness to transmission errors ensured the quality improvement was maintained in real-world cellular environments. EFR set a new benchmark for mobile voice and paved the way for subsequent codecs like AMR, which incorporated EFR as one of its operating modes.