Wideband Multi-Rate Algebraic Code-Excited Linear Prediction

Description

MRWB-ACELP, or Wideband Multi-Rate ACELP, is a speech coding algorithm standardized by 3GPP. It is fundamentally the Algebraic Code-Excited Linear Prediction (ACELP) core of the Adaptive Multi-Rate Wideband (AMR-WB) speech codec, operating with a wider audio bandwidth of 50–7000 Hz compared to narrowband's 300–3400 Hz. The codec operates by analyzing speech signals, extracting parameters related to the vocal tract filter (using linear prediction) and the excitation signal (using an algebraic codebook), and then quantizing these parameters for transmission at variable bit rates.

The architecture of the codec is based on a multi-rate ACELP structure. It supports multiple source codec modes, each corresponding to a specific bit rate. These modes allow the network to command a change in the codec bit rate during an active voice call, a process known as codec mode adaptation or link adaptation. The supported bit rates for MRWB-ACELP (AMR-WB) are 6.60, 8.85, 12.65, 14.25, 15.85, 18.25, 19.85, 23.05, and 23.85 kbit/s. The codec's operation involves framing the input speech (typically 20 ms frames), performing linear predictive analysis to obtain filter coefficients, and then searching fixed and adaptive codebooks to represent the excitation signal that, when passed through the synthesis filter, best matches the original speech.

Its role in the network is to provide high-quality, robust voice service, especially for packet-switched voice over IP (VoIP) in LTE (VoLTE) and 5G (VoNR). The wideband audio provides significantly improved naturalness and intelligibility. The multi-rate capability is key for network efficiency and robustness; the system can lower the bit rate during periods of congestion or poor radio conditions to maintain call continuity with higher error protection, or increase the bit rate under good conditions for superior quality. The codec parameters are packetized into RTP payloads for transport over the IP-based core network.

Purpose & Motivation

MRWB-ACELP was developed to meet the growing demand for high-quality voice services in mobile networks, moving beyond the 'telephone quality' of narrowband codecs. The primary motivation was to provide a more natural and immersive voice experience by doubling the audio bandwidth. This was driven by user expectations set by fixed-line wideband services and the capabilities of new packet-switched network architectures.

It solves the problem of delivering consistent voice quality in a variable wireless environment. The multi-rate nature of the codec addresses the limitations of fixed-rate codecs, which could either waste capacity or drop calls under adverse conditions. By allowing dynamic adaptation, MRWB-ACELP enables optimal trade-offs between voice quality, capacity, and coverage. Its creation was motivated by the evolution towards All-IP networks (IMS), where a efficient, adaptive wideband codec was essential for making VoIP over cellular competitive with, and superior to, traditional circuit-switched voice. It became the cornerstone for HD Voice services in 3GPP networks.

Key Features

• Wideband audio bandwidth (50-7000 Hz) for superior speech naturalness and clarity.
• Multi-rate operation with nine source codec bit rates from 6.6 to 23.85 kbit/s.
• Built on the efficient ACELP (Algebraic Code-Excited Linear Prediction) algorithm.
• Supports in-call codec mode adaptation based on network conditions and commands.
• Provides robust performance against transmission errors through rate adaptation and channel coding.
• Standardized payload format for transport over RTP/UDP/IP, enabling VoIP/VoLTE/VoNR.

Evolution Across Releases

Defining Specifications