I just found out that the XTA DP200 uses DF1760 and PCM 1702 DAC chips.
Has AES input with CS8411.
Was built around 2003? Uses custom programmed DSP chips and external RAM chips.
Sound quality unknown?
Has AES input with CS8411.
Was built around 2003? Uses custom programmed DSP chips and external RAM chips.
Sound quality unknown?
which filters are noiser
Hi,
Interesting read:
....
This data should dispel once and for all the myth that digital filters aren't noisy. With 'only' 24 bits of signal processing, the Direct Form filter is inferior to the analog filter at frequencies below around 300 Hz. At a 10 Hz cutoff frequency, the digital filter's noise level is worse by 34 dB than the analog filter. It's ironic that people often cite digital filters as being of most use in replacing low frequency filters that need high-valued capacitors. We can see, however, that the active filter comprehensively trounces this particular digital filter at low frequencies. The scaling is definitely not working out as you might have expected!.......
link Which filters are noisier – analog or digital? part 2
regards,
Hi,
Interesting read:
....
This data should dispel once and for all the myth that digital filters aren't noisy. With 'only' 24 bits of signal processing, the Direct Form filter is inferior to the analog filter at frequencies below around 300 Hz. At a 10 Hz cutoff frequency, the digital filter's noise level is worse by 34 dB than the analog filter. It's ironic that people often cite digital filters as being of most use in replacing low frequency filters that need high-valued capacitors. We can see, however, that the active filter comprehensively trounces this particular digital filter at low frequencies. The scaling is definitely not working out as you might have expected!.......
link Which filters are noisier – analog or digital? part 2
regards,
maybe you should read a singal processing textbook - wordlength requirements are well known, easily calculated - integer DSP chips use multiple precision - 16 DSP often have 40 bit MAC, many DAW use extended internal wordlength, advertize 64 bit internal wordsize
any designing digital filters should have learned these details - it is a strawaman to say digital processing has noise problems - know the issues, limits and design for them
any designing digital filters should have learned these details - it is a strawaman to say digital processing has noise problems - know the issues, limits and design for them
Last edited:
Xta electronic sound quality
XTA DPxxxx series has a really good reputation, some pro audio manufacturers use this loudspeakers management DSP, like FunktionOne probably one of the best disco/club sound systems around.
I have installed a few of them, with really good results...disco usage of course.
Transformers in/out option expensive but really good, if you stay with analog connections.
I will try a DP426 with a pair of Tannoy Puma soon, I hope....
Will share my finding...
regards
XTA DPxxxx series has a really good reputation, some pro audio manufacturers use this loudspeakers management DSP, like FunktionOne probably one of the best disco/club sound systems around.
I have installed a few of them, with really good results...disco usage of course.
Transformers in/out option expensive but really good, if you stay with analog connections.
I will try a DP426 with a pair of Tannoy Puma soon, I hope....
Will share my finding...
regards
Yeah, that's a nice way of presenting noise that's probably not as commonly used as it should be (appendix D is the other example which springs to mind). I don't think the follow up article on other filter forms was ever written but as others have commented on the article's a little bit disingenuous. While the author's understandably heavily invested in the 24 bit PSoCs most DSPs in DIY audio operates at 28, 32, or 64 bit. Since noise moves by 6dB per bit in fixed point it's often fine to use direct form I everywhere and focus on ordering of filters rather than optimizing filter form based on center frequency. Depending on the levels of optimization available it can also be more computationally efficient to throw bits at direct form I than to switch to, say, state variable.
You might like Mark Allie's deck as well.
Uncorrelated sources add in RMS whether they're analog or digital. With care in analog design noise floors of a few microvolts aren't super hard. Matching is, in my experience, often a 20-80dB larger issue depending on just how much money and effort one's willing to expend on component precision, correction of offsets from bias currents, and so on. So, lower cost and ease of making changes aside, getting exactly the filter coefficient one asks for is IMO a major advantage of digital.
Ah, I wasn't talking 'objective' noise there, rather the subjectively perceived kind. Put another way, stringing together active XOs will reduce dynamics.
Not currently, its one of the things I'd like to get an objective handle on in future. Its most certainly related to how stiff the power supply is, for a given PSRR and loading. Its also related to lack of out-of-band noise - lower in-band IMD products corresponds to better dynamics. Just to give one example of this - putting a steep passive low-pass filter between my DAC's output and subsequent amplification stage notched up the dynamics markedly.
yes XTA has a great repuation. I've heard them quite a few times.
DP426 uses AK4393 DAC chips.
Yeah, there's definitely tradeoffs between signal level (dynamic range over noise floor), drive current in the filters (power supply load variation depending on whether the op amps stay in their class A range or not), and resistor sizing (Johnson and op amp current noise, self heating distortion, and bias current offsets). Signal levels of 1-2V RMS are something of a sweet spot for this, allowing operation in the range of 105dB DNR and -120dB THD if the gain structure supports those levels. Around that operating point it takes a fair number of Sallen-Keys to build up enough of a noise or intermodulation floor to result in something likely to be audible. My experience is managing resistor TCRs is tends to be the hardest part.
In principle it's not hard to work out the DAC stopband depth and analog lowpass image rejection need
Your hunches are completely without merit; and are wrong.
Any complex filter may be implemented with DSP; all results pass through DAC.
Passive XO is just another filter; and relatively simple at that.
I'm not following you - why are my hunches wrong? Nothing you said following your claim supported it that I can see. What am I missing?
'To be audible' - do you mean by applying thresholds of audibility (Fletcher-Munsen et al) or from your own subjective listening evaluations?
With my NOS DACs on RBCD the standard theory says to reach -96dB by 24k1. I haven't got there yet but even -50dB sounds considerably better than no filter at all. What I haven't determined yet is whether aiming for better than -50dB is going to be worth my time, subjectively or even whether -50dB is actually overkill.
Rule of thumb I've found is artifacts which measure -50dBC require a listener who's paying attention to here. -60dBC seems to be around the limit of critical listening. I arrived at this independently but I've come across it other places as well---Ethan Wiener's site and, if memory serves, somewhere in around page 400 or 500 in the Beyond the Ariel thread over in DIYA's multi-way forum. Generally if you design for 80dB down the worst case corners come in better than 60dB and each stage in the playback chain---DAC, output buffer, Sallen-Key, power amp, etc---ends up needing to be around 100dB down to keep the sum down the signal path up to the drivers under 80dB. Stages 120dB down don't add up as fast as 100dB down stages.
Like all rules of thumb this has its limitations, mainly around the ear and brain's ability to pick up on structured artifact below "noise" floors (noise in the canonical sense or other cases such as an intermodulation product floor). It is, however, a place to start which is in good agreement with my subjective experience with quite a few DACs, op amps, preamps, power amps, drivers, and whatnot. Since it's subjective, adjust to taste. But at least it provides something of a framework for discussing the design required for transparency in the pro audio sense of sound in = sound out.
Audiophile senses of "lots of fine detail" or "I like tube amps because the lower feedback reduces cone breakup artifacts relative to solid state", well, those are more complicated. 😉
That Fourier guy has an important point about DAC filters. Things which are abrupt in time are broad in frequency. Things which are abrupt in frequency are broad in time. There's no way around this. So what DAC manufacturers do is present a continuum of options from linear phase brickwall filters with long impulse responses to low group delay slow rolloff filters with short impulse responses. The longer the impulse response the more the DAC's output is blurred in time due to the convolution between the signal and the antialiasing filter's response. The slower the rolloff the greater the amount of aliased energy at higher frequencies propagating down the signal path. The former might bear on what you call dynamics. The latter asks more of the downstream amplifiers and tweeters, raising the intermodulation floor, putting more stress on the supply, and so on.
There's not, to my knowledge, an established and rigorous set of psychoacoustic findings on listener preferences along this continuum. All the results I've come across as well as my own ABX tests align in the direction that getting away from linear phase and brickwall filters generally improves subjective perception of the sound quality. In my experience Cirrus picked up a pretty good balance of filter design tradeoffs when they acquired Crystal. So, if designing one's own antialiasing, there are worse places to start than the slow roll filter responses in their ADC and DAC datasheets.
I've had discussions with Ethan over on another forum and I'd tend to discount almost anything he suggests 😀 However the rest of what you're saying here sounds reasonable (at least as a starting point) - as a target for an IMD products (aka 'dynamic noise' aka 'noise modulation') floor.
The next question that arises is - how to design for such a target when datasheets for opamps don't characterize them for IMD performance? (This being IMD performance with multi-tone stimulus, each tone at -30dBfs or lower, to mimic music) Do you know some way to interpret the customary THD plots to get out the relevant info?
<edit> After posting I realized there is one exception to the rule - the AD8017 has this. And sounds remarkably good 🙂
Attachments
Last edited:
Accurate crossover requires matching of acoustic slopes and phase of driver outputs. With simple passive crossover, and active crossover this is limited by a number of variables. Signal dependent impedance of drivers interacts with crossover components with passive crossover, and level matching requires resistors, or transformers with associated shortcomings.
Component matching is limited with passive components. This limits complexity of filters which may be implemented with analog circuits.
Repeatability, and complexity achievable with DSP is orders of magnitude greater than with analog methods.
Nothing I can see to disagree with there. So back to my original question - why are my hunches wrong?
There are a variety of other amplifiers with both IMD and THD in the datasheets---Analog has a bunch and National included it for the LME497xx series---and for all the ones I've looked at IMD = THD is a good approximation. Same goes for the ones I've measured (though those are all pretty capable parts). This is unsurprising as they're both measures of the same behavior obtained in the same way. Specifically, THD is the special case of n tone IMD where all tones fall at the same frequency. More formally, it's the diagonal of any n dimensional IMD matrix for any arbitrary n.
It's not particularly hard to measure IMD or correlate to it to subjective quality. I apply the 10 tone rake defined for Rnonlin in AudioTester.
For a moment I thought I might have missed something, so I looked again at LME49720 DS - and only found IMD plots vs output voltage into 600R and 10k loads. Nothing like what I was talking about.
So what does this math tell us? That we don't need MTPR measurements any longer because THD tells us all we need to know?
- Status
- Not open for further replies.