An analog crossover at best is equal to a digital filter. Digital filter at 24bits depth has more accuracy than any analog circuit. Processing at 32bits and beyond there is little comparison. Sure, one, two, even four poles per filter is quiet, but it doesn't do good job matching drivers through crossover region.
Here is example of response achievable with DSP:
An externally hosted image should be here but it was not working when we last tested it.
This is individual woofer and tweeter responses:
An externally hosted image should be here but it was not working when we last tested it.
This is reverse null notch when tweeter polarity is reversed:
An externally hosted image should be here but it was not working when we last tested it.
Results at this level are impossible with analog based filters.
I've seen you posting those (or very similar plots) elsewhere. But let's get back to my question shall we - why are my hunches wrong, as you've claimed?
They simply aren't correct, and don't apply. Noise of any competently designed circuit is way below audible threshold. Microphones are noisier element, and effective noise floor of drivers is greatest source of noise due to modulation distortion.
Your hunches disappear in blind testing.
Your hunches disappear in blind testing.
I've probably yet to meet any 'competently designed circuit' then. But I rather suspect my meaning of noise here is not the same as yours.
So far you're still only claiming and not doing any explaining.
Thanks to everyone for posting their thoughts on this topic.
I am now thoroughly convinced that Digital Signal Processing is the way to go for best sounding "audiophile" listening, speaker and room correction.
For a 4 way stereo set up my choices are "Mini DSP 4 x 10", several different brands of pro p.a. units in mono 4 way times two for stereo, "ATX dp448 at $5K and the DEQX 3 way at 3 to 5K.
Opinions are still appreciated!
I am now thoroughly convinced that Digital Signal Processing is the way to go for best sounding "audiophile" listening, speaker and room correction.
For a 4 way stereo set up my choices are "Mini DSP 4 x 10", several different brands of pro p.a. units in mono 4 way times two for stereo, "ATX dp448 at $5K and the DEQX 3 way at 3 to 5K.
Opinions are still appreciated!
Don't forget DIYA also hosts a PC based forum.
I think it says that since you're rejecting the only available bridge between the majority of datasheets and multitone IMD there's probably a fair number of IMD measurements in your future. 😉 If you find parts where THD is not a good predictor of IMD do share; it'll be interesting to see where the approximation breaks down.
If you want "the best", then whatever you choose should use FIR filters. To throw something at you from left field, check out realsoundlab coneq. It can correct your room, and phase. But is only an in line device, not a crossover.
Last edited:
High dollar options are waste of money.
My results in #61 where obtained with <$100 sound card, and freely available software. Caveat is learning curve. With Mini-DSP and it's recent integration with REW, it is much better way to go. From there it is fairly easy to set up complex FIR filters and convolution engine to run correction.
Choice of measurement point(s), and quality measurement microphone with correction data becomes critical to best results. This varies with speaker design.
In my posted results, speaker drivers are spaced <1/4 wavelength of crossover frequency. Good results are possible with somewhat greater spacing, but closer is better.
Had a quick look at the first 50 posts, couldn't find anyone comparing active (digital or not) line level against passive line level. Seemed to be discussions about speaker level passive vs active.
Is there a particular post you could point me to which deals with what I was talking about amongst the 500+?
Would you say a 20dB difference between two-tone IMD and MTPR counted as 'not a good predictor'? This from AD8016 DS :
Attachments
Quite a slog isn't it? You can search that thread for my postings, and address me on those, and think about the posts to which I was responding, and to the posts in response to mine.
At simple level passive crossover, active crossover, and DSP implementation of same simple crossover are theoretically identical. Actual performance isn't. From here as scale of complexity goes up, active circuit and DSP leave passive solution in the dust. As complexity is attempted with active analog to get results as with my previous posting, it is also left in the dust by DSP approach.
At simple level passive crossover, active crossover, and DSP implementation of same simple crossover are theoretically identical. Actual performance isn't. From here as scale of complexity goes up, active circuit and DSP leave passive solution in the dust. As complexity is attempted with active analog to get results as with my previous posting, it is also left in the dust by DSP approach.
Last edited:
Note 1 for MTPR says see figure 48, which is forced air cooling. 😛 If there's a well defined standard for MTPR I'm not aware of it---all descriptions I've come across are essentially conceptual in that a bunch of tones are applied with one missing but the intervals and location of the omitted one aren't defined---but my reading of the datasheet would be that tones from 26kHz to 1.1MHz were applied with a "silence sweep" to find the omission with the highest residual. From figures 11, 12, 14, and 15 this would presumably be one of the tones around 1MHz as that's where the part is least linear.
Table 1 (+-12V) supplies suggests THD into a 100 ohm line at 2V 1MHz would be about -77dB as H2 is -77dB and H3 is sufficiently lower as to be negligible. The -75dB MTPR for that case is 3.2V drive into a 100 ohm line. Given the difference in measurement conditions and the AD8016's fairly complex behavior I'd consider numbers within 2dB of each other to be functionally equivalent. Similarly, the -88dB IMD at 500kHz in this table is reasonable agreement with the H2 slope of figure 12.
In table 2 (+-6V), which is the one you've screenshotted, the proxied THD and IMD values are slightly higher than in table 2 with a larger rise in still small H3. This suggests the part's closer to clipping, which is consistent with the halving of the supply voltage. However, the MTPR in this table is taken at 1.4V rather than 3.2. I think what this is saying is the MTPR test now happens to land on the peak H3 value of figure 22 rather than the below peak condition the table 1 data intercepts in figure 21.
So, while a naïve look at table 2 shows a 20dB difference between IMD and MTPR, all the numbers appear to agree within a couple dB once the part's behavior with supply, frequency, and output level is considered and uncertainties in test methodology taken into account. As Barley's been trying to say, be careful of consulting your intuition---you might get what you asked for. 😉
I follow what Barleywater says. DSP is capable of far higher orders than analogue, with accuracy. Bit of a moot point though, unless your in the camp that believes 48th order filters solve everything....
Me? I'm happy with sallenkey (up to max 8th order), no real benefit in DSP at that level (in my opinion) unless you NEED PEQ, and id rather just get smoother drivers to start with.
Me? I'm happy with sallenkey (up to max 8th order), no real benefit in DSP at that level (in my opinion) unless you NEED PEQ, and id rather just get smoother drivers to start with.
It depends on how hard one's pushing the design. LR4 XO is a good default but I've found improvement with LR6 and, more debatably, LR8. If you're doing a three or four way with an eighth order analogue limit the LR4 highpass and lowpass on the midranges exhaust the available filtering capacity. Smoother drivers tend to be more mechanically damped ones and, even if driver selection requirements allow use of the exceptions which are smooth without impulse response extension from damping, it's my experience ABX shows definite subjective improvement from use of parametric over both graphic and no EQ. Even if it's just one well chosen peaking biquad to fill in a dip. DSP's also pretty useful for prototyping analogue filter banks.
Each of the benefits here is incremental and not a matter of great importance on its own. But, as is so often seems to be the case for high end audio, they accumulate to worthwhile improvement surprisingly quickly. At least if you're trying get that last bit to go from a good system to a great system in a sound in = sound out design.
Each of the benefits here is incremental and not a matter of great importance on its own. But, as is so often seems to be the case for high end audio, they accumulate to worthwhile improvement surprisingly quickly. At least if you're trying get that last bit to go from a good system to a great system in a sound in = sound out design.
Pretty much my understanding yep.
Since hindsight is always 20/20 how about showing your working from the LME49720 DS plots of THD/IMD and estimate from those its MTPR? Based o the work of R.A.Belcher (see his 1978 WW article) my hunch is that MTPR is the best bet we've so far got for bringing perceived SQ and measurement closer to agreement.
My admittedly naive look at your working is that you're rationalizing 😀 If not then the whole DSL industry has wasted its time in developing MTPR as a measure - they could just have consulted you with your crystal ball and saved a pile of hassle.
QED 😀
No, I read it at the time I'm fairly sure - there's a lot of good stuff there but none relevant to what I've said and what you dispute.
OK so scrolling down it seems you joined the party quite late on - post #268 where you talk about current drive vs voltage drive and show a couple of FR plots. Nothing of relevance there. A further heads up?
Plenty of claims there - but none that I can see relevant to my hunches.
Yes - no dispute - DSP and a measurement mic can achieve amazing flatness, if that's what floats your boat.
Indeed - the largest proportion of what Barleywater's been saying has been moot, given that the way I read it, the OP is concerned with how things sound, not technical proficiency per se.
Found post #303 with a whole bunch of claims. Just as on this thread, claims but no explanations.
I'll just address the first claim which is so obviously wrong it doesn't even pass Bruce Schneier's laugh test, to wit -
Amplifiers don't sound like anything, drivers make sound.
No, drivers don't 'make sound' any more than amplifiers do. Drivers vibrate, its our brains that make sound. Sound being a percept.
This follows only if the THD ~= IMD ~= MTPR approximation is sufficiently accurate in all cases, which is unlikely as datasheets are usually expected to have a precision of half a dB or better. A crystal ball is not a maximum likelihood estimator nor is an estimator a measurement.
You're more than smart enough to understand these differences. So if you'd like to continue this conversation stop being unnecessarily rude about it.
- Status
- Not open for further replies.