Method for creating a representation of a calculation result depending linearly on the square a value

Abstract

Transition to the logarithmic range does not require the entire bit range of the result that is linearly dependent on the square of the value. Rather, for a value having x bits, the result can be scaled in such a manner that a representation with less than x bits of the result is sufficient to obtain the logarithmic representation based thereon. The effect of the scaling factor on the resulting logarithmic representation can be cancelled by adding a correction value obtained by using the logarithmic function on the scaling factor to or subtracting it from the scaled logarithmic function without dynamic loss. The invention provides a method and a device for producing a representation of a calculation result that is linearly dependent on the square of a value, thereby allowing to carry out the calculation easily or with little hardware.

Description

technical field [0001] The invention relates to the creation of representations of calculations that are linearly dependent on the square of values, such as the sum of squares of a set of values ​​required in some audio coders in which the frequency line (frequency line) are combined to form a frequency group to use the signal energy of the frequency group to perform further encoding. Background technique [0002] In modern filterbank-based audio coding methods such as MPEG Layer-3 or MPEG AAC, a psychoacoustic model is used in the encoder. In this psychoacoustic model, the entire spectrum of the audio signal transformed into frequency ranges is divided into individual frequency groups, each frequency group having a different width and / or a different number of frequency lines. In order to calculate psychoacoustic hearing thresholds, decide whether center / side stereo encoding should be used, and evaluate and / calculate scaling factors in the quantization module of the audio e...

AI Technical Summary

Problems solved by technology

The content of this document attempts to eliminate the following problems: most input data (ie energy) has an excessively large dynamic range, most fixed-point DSP chips only include a data width of 16 to 24 bits, and MPEG marks require a data width of 34 bits, i.e. 101dB of dynamic range, so energy must first be scaled

However, the downside of this process is that the logarithm calculation is still relatively complicated because the range of values ​​that fit into the 16-bit fixed-point representation of a 16-bit DSP fixed-point processor is only set after the logarithm calculation, while this taking the logarithm is still is performed on energies with large dynamic changes, which results in as many as two logarithmic operations having to be performed for each energy value

Images