Member
Joined 2006
I think one of the unknown is how loop gain over the spectrum varies. If it does not vary exactly the same over the audio spectrum, any variation could effect the total sonic balance. I think we have insufficient evidence how temperature changes this.
Who's "we"? I have plenty of evidence from my day job. I've designed many, many op-amps, bandgaps, and other DC or near-DC (<1-10 MHz) circuits that have been parts of various on-chip building blocks. The general trend is that the loop gain curve of a Miller-compensated op-amp will shift up/down in magnitude with temperature (high temp -> lower gain). The dominant pole will shift in frequency depending slightly on temperature, but more so on process variation. I have never released an op-amp design to production that showed any meaningful difference in closed loop performance across temperature, supply voltage, or process variation. This was the design philosophy at National Semiconductor - at least as far as I experienced it. This is also the case with the Modulus-86 and Parallel-86.
You can easily verify my claims above by setting up a simulation in your favorite circuit simulator. Design an opamp with a handful of transistors and characterize its open loop gain vs power supply voltage, temperature, and component variation.
Exactly.
Tom
We in this thread does not show whether temperature effects the loop gain the same way at all audio frequencies. At least I see lots of talk but no data. Not that I think this is an important issue because there is not much a chip user can do cost effectively to counter the variation.
The problem with simulation is that SPICE simulation is not precise enough. It is great for exploring ideas, but if you want to do serious work, lots of modelling needs to be customised to a specific manufacturing process. Even layout effects performance as well. You just cannot use a discrete SPICE model an ensure that it is the same as the chip.
Quite: Since LTspice became more generally available many forget (or didn't know) quite how much time those in the semi industry spend getting their SPICE models accurate for the process. Given the cost of a reticule set you can see why.
Check the heading of the topic. This is a build thread to help people build their Mod-86, not some deranged willy waving on strange beliefs and half truths. Go find somewhere else to peddle your musings please.
I work with the miniDSP for loudspeaker development and it is a very nice piece of kit.
However, its performance does not come close to that of the Modulus 86. For that, you will have to build your own analog xover. Linkwitz has good information on his site and I think also sells boards. Also check out Elliot Sound Products.
Member
Joined 2006
Yes, THD and noise. Between miniDSP and a well made analog xover using LM 4562, there is about 20 dB difference. Not that miniDSP is bad, not at all, it is very good. But not the same ballpark as the Modulus86.
I have measured these differences time and again. These are Q&D (quick and dirty) measurements comparing the mid section of a xover.
This is the miniDSP
As compared to the analog version
Please notice that for the analog version, I had to use a 15 dB scale compared to the 10 dB scale for the miniDSP.
A more recent analog xover I assembled has even better performance (because of better lay out etc.)
I have measured these differences time and again. These are Q&D (quick and dirty) measurements comparing the mid section of a xover.
This is the miniDSP
As compared to the analog version
Please notice that for the analog version, I had to use a 15 dB scale compared to the 10 dB scale for the miniDSP.
A more recent analog xover I assembled has even better performance (because of better lay out etc.)
Given the DSPs available with high-precision math (32- or 64-bit) there's really no technical reason a DSP filter can't outperform an analog filter. However, as with any precision design, one has to pay attention to the architecture and design. I'm not familiar with the architecture or code base of the MiniDSP, so I can't say specifically if has the capability to outperform analog filters. Above plots make me wonder, though...
Even analog filters have their limitations. You really have to pay attention during the design phase to ensure that you get the stop band attenuation you need. I ran into that during the design phase of the DC servo in Modulus-86 Rev. 2.0 (there! back on topic...
) I suggest reading Bonnie Baker's note on the topic: http://www.ti.com/lit/an/slyt306/slyt306.pdfTom
Last edited:
For precision circuits, the layout is part of the circuit. I was able to get a 10 dB improvement in mains hum (from about -110 dBV to -120 dBV) by the removal of about 4 mm of PCB trace when I updated MOD86 from R1.0 to R2.0.
Tom
Member
Joined 2006
Tom, the slopes of my analog filters dive right into the noise, but you need good caps (polyprop) and short traces (induction) to do so. Will read Bonnie Bakers notes later today, thanks for the hint.
Back to the thread's topic: as you mention, getting the grounding right, as well as other issues connected to layout, is essential to eke out the last bit of performance from a design. Hence I applaud your efforts to carry the LM3886 to stratospheric levels.
CFT, I imagine I do hear a difference, but it is psychologically impossible to think otherwise without methodical ABX-ing.
The main reasons to not use DSP xover are not audibility. The first is latency; speakers with DSP processing are too slow to use for live monitoring and can only be mated to video with provisions for delay. The second is gain structure. DSP processing is nice if you have control over the gain structure, with active speakers, you don't.
Back to the thread's topic: as you mention, getting the grounding right, as well as other issues connected to layout, is essential to eke out the last bit of performance from a design. Hence I applaud your efforts to carry the LM3886 to stratospheric levels.
CFT, I imagine I do hear a difference, but it is psychologically impossible to think otherwise without methodical ABX-ing.
The main reasons to not use DSP xover are not audibility. The first is latency; speakers with DSP processing are too slow to use for live monitoring and can only be mated to video with provisions for delay. The second is gain structure. DSP processing is nice if you have control over the gain structure, with active speakers, you don't.