Modified Intermediate Reference System

Description

The Modified Intermediate Reference System (M-IRS) is a crucial component within the 3GPP standardization framework for speech quality assessment. It is not a physical device but a precisely defined mathematical model that simulates the frequency response of a specific, historically significant telephone handset. This model is implemented in software and used as a reference filter in objective measurement algorithms like Perceptual Evaluation of Speech Quality (PESQ) and its successor, POLQA. When evaluating a speech codec or network path, the original clean speech signal is first filtered through the M-IRS model to simulate the effect of a typical receiving terminal. This 'modified' signal then serves as the reference against which the degraded output signal (after encoding, transmission, and decoding) is compared. The core architecture of M-IRS is its digital filter coefficients, which are specified in 3GPP TS 46.085. These coefficients define a transfer function that accurately replicates the amplitude-frequency response of the Intermediate Reference System (IRS) defined by the ITU-T, but with modifications to better represent certain regional telephone characteristics. The system works by applying this linear, time-invariant filter to the audio signal, attenuating or emphasizing specific frequency bands to mimic the non-flat response of a real electro-acoustic transducer. Its role is to ensure that quality measurements are not just about the digital codec's performance in isolation, but about the end-to-end auditory experience as it would be perceived through a typical handset, thereby grounding laboratory tests in practical, real-world listening conditions. By providing this fixed, unchanging reference point, M-IRS enables fair and consistent comparisons between different codec implementations, network equipment from various vendors, and the performance of a system across different 3GPP releases. It is a foundational element for quality assurance, codec selection, and network optimization processes.

Purpose & Motivation

The M-IRS was created to standardize and objectify the process of speech quality evaluation in digital telecommunication systems. Prior to its adoption, subjective listening tests were the primary method for assessing codec quality, which were time-consuming, expensive, and prone to variability. The industry needed a reliable, repeatable, and automated objective method. The purpose of M-IRS is to serve as a controlled, reproducible 'listening terminal' within these objective measurement tools. It solves the problem of inconsistent test conditions by defining exactly how the reference signal should be acoustically shaped before comparison. This ensures that quality scores for a GSM Enhanced Full Rate (EFR) codec tested in Europe are directly comparable to those for an EVS codec tested in Asia, as both are judged against the same simulated receiver characteristic. Historically, different regions used slightly different telephone handset frequency responses (e.g., North American vs. European characteristics). The M-IRS, based on the ITU-T IRS but with modifications, provided a single, globally accepted compromise model for 3GPP's mobile context. Its creation was motivated by the need for efficient codec development and network benchmarking during the rapid evolution from 2G to 3G and beyond, where proving superior or equivalent voice quality was a key competitive and regulatory requirement.

Key Features

• Standardized digital filter coefficients defined in TS 46.085
• Simulates the frequency response of a typical telephone handset
• Provides a consistent reference for objective speech quality measurements (e.g., PESQ, POLQA)
• Enables reproducible and vendor-neutral testing of speech codecs
• Integral part of 3GPP codec performance verification and compliance testing
• Foundational for end-to-end voice quality assessment in mobile networks

Evolution Across Releases

Defining Specifications