Overview

Originally developed to be used in GSM by the ETSI, the Adaptive Multi-Rate (AMR) speech codec [TS 26.071] was approved within the 3GPP forum in 1999 to be mandatory for circuit- and packet-switched speech in UMTS networks. An AMR speech codec adapts the error protection level to the local radio channel and traffic conditions so that it always selects the optimum channel and codec mode to deliver the best combination of speech quality and system capacity. AMR uses Multi-Rate Algebraic Code Excited Linear Prediction (MR-ACELP) scheme based on two different synthesis filters. It converts a narrowband speech signal (from 300 to 3,400 Hz) to 13-bit uniform Pulse Coded Modulated (PCM) samples with 8 kHz sample rate. This leads to 20 ms AMR frames consisting of 160 encoded speech samples. This means that the codec can switch mode, i.e. source bite rate, every 20 ms. AMR has 8 coded modes in UMTS systems, whereas in GSM AMR uses either 6 or 8 modes. The eight source rates vary from 4.75 to 12.2 kbps. It also contains a low rate encoding mode, called Silence Descriptor (SID), which operates at 1.8 kbps to produce background noise and a non-transmission mode.

The AMR codec dynamically adapts its error protection level to the channel error conditions. For instance, lower speech coding bit rate and more error protection schemes are used in bad channel conditions. This principle is illustrated in Figure A1.1 where AMR strives to change to the best curve...