Considering modification to the timing of the audio processor leads to a change in sound, thought was given to the effect of processor-induced IMD within the codec. The following simple test was crafted to observe the effects of IMD through a codec.

Figure 1. The test setup used to analyze low bit-rate audio.

Figure 1 illustrates the test setup. A multi-tone sinewave generator creates the source signals to stress the audio processor and codec. Frequencies were set to 400Hz and 11.5kHz. The output from the audio processor was routed in two directions — to the input of a multi-channel spectrum analyzer, and to the input of an HD Radio encoder. The encoder was routed directly to a corresponding decoder, and its output was connected to the other input of the spectrum analyzer.

The objective of this test was to observe whether or not any part of the dynamics function will generate distortion via the codec. The audio processor employed for the test was designed to condition audio in a coded environment. The back-end processing utilized look-ahead limiting in place of hard limiting/clipping. This reduced THD components in the codec and eliminated aliasing in the system. Tone bursts of the twin tones were used, as this would simulate the effects of transient activity in the source signal, as well as activate the fast-limiting functions in the audio processor.

Figure 2. A spectral display of the 400Hz and 11.5kHz tone bursts at the output of the audio processor.
Click image to enlarge.

Figure 2 is the spectral illustration of the tone bursts at the output of the audio processor. The twin-tones appear as would be expected. This is also the result when observed at the output of the codec when steady-state tones are passed through the processor and codec together.

Figure 3 illustrates the output of the codec's decoder. Notice the significant spectra around the upper frequency of 11.5kHz. Further investigation of the situation revealed that the transient activity upset the encoder and caused added modulation in the upper frequency domain. This is what was causing the added ghost-like product heard prior. Is this possibly the effect of the SBR function becoming upset at transient information? This diagnosis is subject for a deeper discussion.

Figure 3. The output of the audio decoder.
Click image to enlarge.

The rigor of this test exhibited what appeared to be severe IMD in the signal. While broadcast source material does not contain transient twin-tones, it does contain plenty of dynamically transient signals within this frequency range. The extent of this added IMD is dependent upon the transients embedded in the source material. Additionally, fast-limiting time constants in the audio processor are capable of exaggerating, and/or creating this problem.

LoIMD

As with most discoveries, there's an answer. In the above case, further study of the presence and high-frequency limiting algorithms yielded a method to reduce processor induced IMD. Utilizing a proprietary new function known as LoIMD, the algorithm is capable of providing fast-limiting to control transients, yet without agitating the encoder. When normal source content material is applied, the audio through the entire coded system is devoid of the ghost-like annoyances that were mentioned earlier.

The LoIMD function modifies the control function within a dynamics algorithm. Through internal analysis of the incoming dynamics, and IMD characteristics, the architecture of the control method is rearranged to provide a control signal that reduces, and sometimes eliminates IMD in the processed signal. The sonic result is cleaner sound for a given amount of gain control.