Digital Signal Processors (DSP) and IIR Filters
It’s been about 5 years since I published a couple of articles on the differences between Digital Signal Processors (DSP). You can find those articles on the SynAudCon website. I had to revisit the topic on a recent loudspeaker measurement project.
The back story is that a loudspeaker manufacturer sent me a two-way loudspeaker for testing, along with the required settings (IIR filters) for the Digital Signal Processor. The resultant response did not look as expected, so we investigated and found the DSP to be the culprit. I had the manufacturer send me a measured IR of their Digital Signal Processor, and I compared the frequency response magnitude to two DSPs that I have in the measurement rack. Figure one shows the comparison.
Figure 1 – A comparison of three popular DSPs with the same settings. Click to enlarge.
So, Digital Signal Processors aren’t any closer now than they were when I did the comparison articles 5-years ago. A few thoughts:
- No one is necessarily right or wrong regarding the response shape of an IIR filter. There is no universally accepted definition for the relationship between the filter Q and bandwidth.
- The differences are not a problem when the loudspeaker response is actually being measured, since the same target curve can be reached with any Digital Signal Processors. An example is when equalizing a sound system.
- The differences are a problem when filters are numerically dialed into a DSP and the resultant response is not verified, or when one make/model Digital Signal Processor is used to prototype the loudspeaker and a different make/model DSP is used to deploy the filters. I run into this quite often when producing loudspeaker data files for room modeling programs.
- If a loudspeaker really, really needs a specific Digital Signal Processor curve (not unusual), the quick way to get it is to use an FIR filter. These are supported by a growing number of Digital Signal Processors. The response of an FIR is not dependent on the DSP, so long as the required tap length is supported. I can load an FIR into my measurement program to process the stimulus. No DSP needed.
- This loudspeaker would have “worked” with any of these Digital Signal Processors, but the results would sound slightly different. Such differences are often attributed to the wrong cause, such as the digital resolution, the “quality” of the product, or the oxygen content of the copper cables.
Technological advancements in audio have not eliminated the need to measure. pb