DSP Comparison – If They Sound Different, They Are Different… (Part 1)
By Pat Brown
How do DSPs compare? Pat Brown decided to test several DSPs and publish the measured data.
It is a buyer’s market when it comes to digital signal processors (DSP). It seems like everyone has one, and that they all do about the same thing. It is tempting to believe that all of these devices will produce the same response if programmed with the same settings. Many users claim to hear a difference between brands, a claim which we do not dispute. But some of these differences are likely due to simple variances in the way that the filters are programmed.
We decided to test the theory by measuring some of the DSPs that we have in the shop. Jeremy Johnston performed the tests and graphed the results. The drive functions are typical of what might be required for a multi-way, bi-amplified loudspeaker.
Let me say up front that we are not trying to prove which unit is “right” and which might be “wrong.” While electronic filter parameters such as frequency and level can be defined with great accuracy, there can be ambiguity regarding filter Q and slope. We simply want to show that there is a difference, and that the differences are large enough to swing the results of an A/B comparison.
Why does any of this matter? Well, sometimes it doesn’t. For some loudspeaker designs and applications, “close” is good enough and there is no great need for high accuracy in the settings. If the user is basing the settings on an acoustic measurement, the filter placements and shapes can be massaged to produce the desired acoustic response, and the numbers that describe the filter behavior are not of particular interest. But if they are basing the DSP settings on a list of input parameters provided by a manufacturer, and they are using a different make/model DSP than was used to produce the settings, there will likely be a significant difference between the response promised and the one that they get. For phase-sensitive loudspeaker systems, especially high-Q line array elements, the differences can even have a significant effect on the radiation pattern of the array.
The bottom line is if your loudspeaker application requires a precise drive function to achieve a target response curve, a list of numerical settings to enter into the DSP may be insufficient for producing the desired results. Instead have the manufacturer provide a graph of the required magnitude and phase response (the transfer function). The Common Loudspeaker File Format provides the means for inclusion of high resolution transfer function plots for line array elements and other loudspeakers that might require them. The measured balloons were gathered using these drive functions. If the manufacturer has tight quality control regarding transducer performance, you will likely get an acoustic response similar to the published specification – but that’s another subject.
We hope that you find this study interesting. We will follow it with Part 2 in our next issue. pb
Technical Overview of Testsby Jeremy Johnston
Each DSP was setup as a bi-amp, 1-input, 2-output loudspeaker controller. The input voltage was set as nearly as possible to 0.7Vrms using the pink noise and pink sweep generator in the measurement platform (SmaartLive™ in this case). I selected a crossover frequency of 1000Hz for the 4th order, Linkwitz-Riley crossover in each DSP. The Lake Contour was selected as the baseline curve. All devices were used well within their linear operating range and no dynamics processing was enabled for these tests. The latency of each DSP was different; I used the “Auto Small” delay locator function in SmaartLive to find the peak of the impulse response on each output of each DSP. This allowed the phase curves to be compared, but the latency of each DSP is not compared here. No delay was used on the high frequency output of the DSPs, though that is a common technique used to compensate for the physical offset of drivers in a bi-amplified loudspeaker. All plots were made from data exported from SmaartLive into a 3rd-party graphing program.
For the each low pass section a 4th order, Butterworth high-pass filter was set at 30Hz and a single parametric filter was inserted. The parametric filter was set to 250Hz, -4dB, with Bandwidth = 0.5 oct. (Q=2.871). For the high pass section, two parametric filters and a high-frequency shelving filter were inserted. The first parametric filter in the high pass section was set to 2500Hz, -3dB, with Bandwidth=0.7 oct. (Q=2.041). The second parametric filter was set to 5000Hz, -6dB, with Bandwidth=0.3 oct. (Q=4.8). The shelving filter in the high pass section was set to 12,000Hz, +4dB. Each DSP had a different interface to contend with and this was decidedly apparent in setting the shelving filter settings (see below).
The low-pass output level of each DSP was set to show no gain in the transfer function. The high-pass output level was set -10dB below the low-pass output. It should be noted that the numbers entered into each DSP were as close as possible to one another given the differences in the user interfaces of each unit. The dbx DrPA and BSS OmniDrive, for example, didn’t allow direct entry of the same Bandwidth numbers as the other units. These units were matched as closely as possible to the numbers selected for this comparison.
As mentioned in the previous DSP setup description, each DSP user interface was unique. This was particularly evident when setting the high-frequency shelving filter that was inserted in the high pass section of the DSPs. The capabilities of each DSP ranged from “no shelving filter” function on the BSS OmniDrive FDS-388 to a shelving filter function that allowed low, mid and high cut or boost on the QSC Basis. For the purpose of this comparison the shelving filters, where available, were each set with the same corner frequency and gain, but the shape of each filter was very different. This shows up well in the transfer function plots shown in the comparison. This would of course produce significant audible differences during an A/B comparison of any of the units. The differences in filter topology and shape are further reason to request transfer function details from the loudspeaker manufacturer when specific parameters are required for specified loudspeaker performance.