Digital transmission of audio signals requires very low encoding/decoding delays. The Ultra Low Delay Codec, developed on behalf of the Fraunhofer Institute for Integrated Circuits IIS, obtains a very low coding delay with basically no loss of compression performance.

Applications which use digital transmission of audio signals, like digital microphones in live or studio settings, in-ear monitoring for musicians, wireless digital transmission to loudspeakers, or musician playing together remotely, require minimal delay. In contrast to standard perceptual audio coding schemes (e.g. MPEG-2 AAC or MPEG-4 AAC Low Delay), the Ultra Low Delay audio coding technique uses a backward predictive coding procedure instead of sub-band coding thus eliminating the inherent algorithmic delay of sub-band coding. With predictive coding it is possible to obtain a very small encoding/decoding delay with basically no loss of compression performance. Hence it is not necessary to choose between delay and the resulting audio quality.


Since the Ultra Low Delay coding scheme uses a perceptual model which is based on a sub-band decomposition of the audio signal, the problem of applying the output of a psycho-acoustic model to predictive coding has to be solved. This is done by separating the application of psycho-acoustics (the irrelevance reduction) from the redundancy reduction.The output of the psycho-acoustic model controls an adaptive pre-filter in such a way, that not audible information contained within the signal is eliminated. The post-filter in the decoder has a frequency response which is the inverse of the pre-filter, hence a near-transparent reconstruction of the input signal can be achieved.After pre-filtering, a simple uniform quantizer is applied followed by a lossless predictive coder. The latter is based on backward prediction, like the so-called LMS algorithm to obtain a minimum algorithmic delay. The prediction error is then entropy coded with a Huffman or an arithmetic coder.The resulting algorithmic encoding/decoding delay is about 6 ms at 44 kHz sampling rate, and still only about 10 ms, including hardware delays in a current implementation.

Currently, the research on Ultra Low Delay coding is continued at the Technische Universität Ilmenau.