How to calculate the  real bandwidth  for an audio codec Craig Southeren Last updated 17 May 2004

Contents

Introduction

Calculating the actual bit rate

Comparing codecs

Change history

Introduction

Users and developers often compare audio codecs on the basis of the codec raw bit-rate. This is does not yield results that are useful in the real world, as the amount of data in the encapsulating UDP[2] and RTP[1] headers can often be larger than the audio data payload.

Calculating the actual bit rate

As an example, consider a stream of G.723.1 packets each containing a 24 byte encoded audio frame. The frame rate for this codec is 30ms, yielding an audio bandwidth of:

( 24 * 8 ) / ( 30 * 0.001) = 6400 bits per second

However, each of these frames is accompanied by an IP header of 20 bytes, a UDP header of 8 bytes, and RTP header of 12 bytes (assuming no CSRC field) for a minimum overhead per frame of 40 bytes. This gives an actual bandwidth of:

( (40 + 24) * 8 ) / ( 30 * 0.001) = 17066 bits per second

which is nearly three times the "headline" bandwidth of the underlying audio codec. This simple calculation ignored additional bits of overhead added by the transport layer, such as Ethernet headers etc, which will make the overheard "on the wire" even higher.

This disparity can be reduced, at the cost of latency, by putting more than one frame of audio into a single UDP packet. If we put three G.723.1 audio frames into a single packet sent every 90ms, the real bit rate is as follows:

( (40 + (3 * 24) ) * 8 ) / ( 3 * (30 * 0.001) ) = 9956 bits per second

This is only 1.5 times the "headline" bandwidth, as the 40 bytes RTP/UDP overhead is now smaller than the payload size of 72 bytes.

Comparing codecs

The following tables gives the actual bandwidth for many of the popular audio codecs in use. This data was calculated using an Excel spreadsheet that can be downloaded