Adaptive Multi-Rate Narrowband

Description

AMR-NB is a multi-rate speech codec that employs Algebraic Code Excited Linear Prediction (ACELP) technology with eight discrete bit rates: 4.75, 5.15, 5.90, 6.70, 7.40, 7.95, 10.20, and 12.20 kbps. The codec operates on 20 ms speech frames, each containing 160 samples at the 8 kHz sampling rate. Each frame is processed through a series of signal processing stages including linear predictive coding (LPC) analysis, pitch analysis, and algebraic codebook search to generate the compressed speech representation.

The architecture consists of an encoder that analyzes speech signals to extract parameters including LPC coefficients, adaptive codebook parameters (pitch lag and gain), and fixed codebook parameters (algebraic codebook indices and gains). The decoder reconstructs speech using these parameters through a synthesis filter that combines the excitation from both codebooks. The codec includes voice activity detection (VAD) and discontinuous transmission (DTX) mechanisms to reduce transmission during silence periods, significantly conserving network resources.

AMR-NB implements source-controlled rate adaptation where the network can command the mobile station to switch between different codec modes based on channel quality measurements. This adaptation occurs at the radio resource control layer, with mode changes typically happening at frame boundaries. The codec includes comfort noise generation (CNG) during DTX periods to maintain natural-sounding silence and prevent the perception of a dropped call. Error concealment mechanisms help maintain acceptable voice quality during frame erasures by extrapolating parameters from previous good frames.

In the network architecture, AMR-NB operates primarily in the user plane between the mobile terminal and the core network's media gateway. For circuit-switched voice, the codec processes speech in the terminal's baseband processor and transmits it over the air interface. In packet-switched implementations like VoLTE, AMR-NB frames are encapsulated in RTP/UDP/IP packets for transmission over the IP multimedia subsystem (IMS). The codec's interoperability with legacy systems is maintained through transcoding functions in media gateways when connecting to PSTN or other networks.

Purpose & Motivation

AMR-NB was developed to address the limitations of earlier GSM speech codecs (Full Rate, Half Rate, and Enhanced Full Rate) which operated at fixed bit rates and couldn't adapt to changing radio conditions. Previous codecs either provided good quality but inefficient spectrum usage or conserved bandwidth at the expense of voice quality. The fixed-rate approach meant networks had to be dimensioned for worst-case conditions, leading to inefficient resource utilization.

The primary motivation was to create a codec that could maintain consistent voice quality across varying channel conditions while optimizing spectral efficiency. By allowing dynamic rate switching, AMR-NB enables networks to trade off between voice quality and channel capacity in real-time. This adaptability was particularly important as cellular networks expanded into areas with challenging radio environments and as subscriber density increased, putting pressure on limited spectrum resources.

AMR-NB also addressed the need for a universal speech codec that could work across different network generations and facilitate international roaming. Its design considered backward compatibility with existing infrastructure while providing a migration path to more efficient voice services. The codec's multiple rate options allowed operators to implement different quality-of-service policies and optimize their networks based on specific deployment scenarios and business requirements.

Key Features

• Eight discrete bit rates from 4.75 to 12.2 kbps for flexible quality-bandwidth tradeoffs
• Algebraic Code Excited Linear Prediction (ACELP) coding technology for high speech quality
• Dynamic rate adaptation based on channel conditions controlled by network commands
• Voice Activity Detection (VAD) and Discontinuous Transmission (DTX) for silence compression
• Comfort Noise Generation (CNG) during silent periods to maintain natural conversation flow
• Robust error concealment mechanisms for frame erasure conditions

Evolution Across Releases

Defining Specifications