1. Introduction
This document describes the error generation software for core experiments on JPEG-2000 error
resilience. A similar program was provided for core experiments on MPEG-4 Video/Audio error resilience
by MPEG96/M1492[2], assuming simple models of error patterns supplied to the coding layer from the
multiplexing layer and the physical transmission layer. It would be a great benefit to use the error
conditions identical with MPEG-4 Video/Audio as a reasonable shortcut.
In this document, two types of modeled error pattern generation programs are described, namely erand
and gilbert. erand creates a bit error sequence corresponding to random errors, while gilbert to burst
errors. These error patterns are very frequently observed in the actual mobile radio channels and base band
modem transmission. Their source programs are attached with the document.
Note that so-called packet loss program is not introduced here, since it requires the frame structure of
coded bitstream and interface definition between coding layer and underlying layers. That would be
provided later when these specifications are given.
First of all, the background that the artificial simple error models was introduced in MPEG-4
Video is summarized in section 2, by referring WG11/N1223. [1] Detailed software description and
its usage for each error model is described in section 3.
2. Background
As described in [1], there were exhaustive discussions on what error patterns should be used in the core
experiments in MPEG-4 Video error resilience. The major points were as follows:
� realistic errors collected in the field, or simplified model-based errors
� what type of channels should be considered
In the absence of underlying layer’s definition, in particular, the multiplexing layer, and in order to
avoid to tune a specific channel, the MPEG-4 error resilience video people concluded to employ simple
models of error patterns as typical ones supplied to the source coding layer. They consist of Random
errors, Burst errors with two-state Markov model, and Random packet loss with a variety of length. These
models cover a wide range of actual channels, for instance, wireless transmission systems like
GSM/PDC/USDC/DECT/PHS/(IMT-2000), internet, base-band modems, and so on. Detailed
specifications, such as bit error rate, burst error length, and packet loss length were decided later as written
in [4]. The error generation programs were provided by the authors, according to the agreement.
The above approach is thought as the first phase. The second phase should consider more realistic error
conditions, taking actual error patterns obtained in transmission channels as well as the multiplexing layer
into account. This will be done in the verification test scheduled next March.
3. Software description
The software can generate three classes of error conditions, random error and burst error with 2 state
Markov model (or Gilbert model). The name of the programs are erand and gilbert, respectively.
We briefly describe the usage of these two programs below. The terms <Input file> and <Output file>
in the following description means the name of input sequence file and output sequence file respectively.
You can use ‘-‘ for the filename, which means standard input/output. All the sequence files processed with
these programs must be in binary form. The transmitted bits should be packed into octets so that their MSB
(most significant bit) is transmitted first.
3.1. Random error (erand)
This software injects random bit error under specified bit error rate. The usage of this program is:
