Hypex DSP module(s)

May I suggest that the remote control be compatible with one (or multiple) of the formats commonly used?

I.e. the Philips RC-5 or NEC's format. Ideally, I would like to be able to use something like an Apple Remote to control my preamp as these remotes are easy to decode (uses the NEC format) and are incredibly inexpensive ($20). The new model is very nicely designed.

~Tom
 
hi Jan Willem
Untitled Document
what price do you expect the DLCP will cost in €?

Regards

For the complete DLCP platform the price is EUR 825,00 ex. VAT and shipping.

The complete platform is; 6 channel DSP board, input/output board with connectors, LCD screen with control buttons and cables.

At the moment we have a beta program running, beta tester can get some discount. If your interested to be a beta tester please contact Jan-Willem, jan-willem (at) hypex (dot) nl

Best regards,

Jan-Peter
 
Practical Design

For the complete DLCP platform the price is EUR 825,00 ex. VAT and shipping.

The complete platform is; 6 channel DSP board, input/output board with connectors, LCD screen with control buttons and cables.

At the moment we have a beta program running, beta tester can get some discount. If your interested to be a beta tester please contact Jan-Willem, jan-willem (at) hypex (dot) nl

Best regards,

Jan-Peter

That DLCP platform looks impressive. Nevertheless, I advise eliminating the LCD display, as well as any buttons (aside from the obvious necessities, such as the power button), to reduce the cost of the finished product (particularly since config is accessible via USB/other on-board connectors).

I can appreciate the impetus to provide a robust UI, but costs should be reduced whenever possible (without degrading the quality of the product, of course). Proper config of a system is done once--set-it-and-forget-it.
 
For the complete DLCP platform the price is EUR 825,00 ex. VAT and shipping.

The complete platform is; 6 channel DSP board, input/output board with connectors, LCD screen with control buttons and cables.

At the moment we have a beta program running, beta tester can get some discount. If your interested to be a beta tester please contact Jan-Willem, jan-willem (at) hypex (dot) nl

Best regards,

Jan-Peter

I just noticed that earlier in this thread you had noted that the DSP module will be available without the LCD and other extras. Please disregard my previous post.
 
For the complete DLCP platform the price is EUR 825,00 ex. VAT and shipping.

The complete platform is; 6 channel DSP board, input/output board with connectors, LCD screen with control buttons and cables.

At the moment we have a beta program running, beta tester can get some discount. If your interested to be a beta tester please contact Jan-Willem, jan-willem (at) hypex (dot) nl

Best regards,

Jan-Peter

can I trade my two AS100.2 amps in for this DSP and beta test it??

I would test it against my MiniDSP and DCX2496.
 
Will the new DSP have phase correction ?

It doesn't have FIR filtering but,

Copied from the HFD FAQ:
The biquad filters are all minimum phase, which means that phase is directly linked to magnitude response in exactly the same way as an analogue filter. Since individual speaker drivers are minimum phase too, EQ'ing them flat also equalises the phase response.



Regards,
Jan-Willem
 
Copied from the HFD FAQ:
The biquad filters are all minimum phase, which means that phase is directly linked to magnitude response in exactly the same way as an analogue filter. Since individual speaker drivers are minimum phase too, EQ'ing them flat also equalises the phase response.

Jan-Willem,

Sorry to but in but this is not true. Take two idealistic minimum phase drivers, use a 1 kHz LR4 (for example) and look at the result. You will see the phase wrapping at 1 kHz ! and you will have the corresponding group delay either.

An overall phase correction through a FIR should be able to give you a flat phase all over.

Jean Claude
 
Let's not confuse a few things.

There's the phase response of the driver itself. This is automatically taken care of by minimum phase filters. That is to say, correct a driver's magnitude response flat and the phase response is flat too (apart from time-of-flight delay of course). You most definitely do not need FIR filters to iron out phase errors of the drivers since correcting the corresponding magnitude response errors does the phase bit for free.

We don't just correct the driver flat however. In the end you want e.g. the LF output (=combination of filter and driver) to exhibit an LR4 lowpass response. The phase response now becomes that of an LR4 lowpass filter. Same with the top end. And when you add the two (duly delay corrected) the sum is a second order allpass filter.

One could envisage correcting the phase of the sum response. A straight IIR filter won't cut it so you would need a FIR filter. This filter would be inserted in the input, i.e. be common to both LF and HF. Its impulse response would be the reverse of that of the idealised second order all-pass filter.

I've done this in a commercial product but I was surprised at how vanishingly small the audible effect was. It took me quite a bit of time to find music that would actually make it detectable at all. Linear phase is hugely overrated. What truly matters is precise phase matching of the crossover filter. IIR and delay correction are the only tools you need for that.

Furthermore the length of the filter is determined by the crossover frequency. For 200Hz it would be much longer than the DSP on the DLCP will handle, requiring a more high-end DSP chip just for a bit of phase correction. Given the above I'd rather see the money go to the analogue circuits, which is what drove the design of this platform.

It is very hard to escape the conclusion that adding the DSP capacity needed to execute this filter would add more cost than value while tooling up the PC software to design alternative tapped delay + IIR filters will take more work than it's worth.

In short, there is no correction for overall phase on the DLCP. The correction of driver phase anomalies comes as a matter of course.
 
Let's not confuse a few things.

Hi Bruno,
English is not not my native language so maybe I was not clear in my last message. I completely agree with your driver analysis.

One could envisage correcting the phase of the sum response. A straight IIR filter won't cut it so you would need a FIR filter. This filter would be inserted in the input, i.e. be common to both LF and HF. Its impulse response would be the reverse of that of the idealised second order all-pass filter.

Yes this is what I tried to say with an overall FIR phase correction and of course it should be done (by anticipation) in the input (s).

If I talked about this is that as far as I remember it was mentioned in this thread that FIR will be implemented by an additional board

I perfectly understand the tribute to pay (latency and DSP power) I thought that using FIR for phase correction only would reduce the number of tap needed at a more reasonable level (no need for fine resolution at low freq).

I've done this in a commercial product but I was surprised at how vanishingly small the audible effect was.

Well I can't comment this myself, having yet to play to play with FIR filters !
 
If I talked about this is that as far as I remember it was mentioned in this thread that FIR will be implemented by an additional board
Well there are always vague plans to add it at some later stage (hence the empty PCB footprint for an extra DSP) but I have to be realistic and not promise anything until we're sure it's coming.

I perfectly understand the tribute to pay (latency and DSP power) I thought that using FIR for phase correction only would reduce the number of tap needed at a more reasonable level (no need for fine resolution at low freq).
That is correct. A FIR filter that only does overall phase correction is going to be shorter than one that equalises the whole thing. But it still is going to be quite a lot if you want to phase-equalize a crossover filter of, for example, 200Hz. A typical design with nice low pre-echo would be 3800 taps long at 192kHz. So for two channels that's around 1.5GMAC/s.
This means that the straight FIR is out of the question for a reasonably priced DSP. An alternative that greatly reduces processing combines a tapped delay line and several IIR filters to approximate the inverse all-pass function. This is much more economical but it means a lot of work for the poor bloke who has to write the design software. This poor bloke is called Jan-Willem so go easy on him ;)
 
Well there are always vague plans to add it at some later stage (hence the empty PCB footprint for an extra DSP) but I have to be realistic and not promise anything until we're sure it's coming.

So I understand that FIR is put on a side at least for a while

3800 taps long at 192kHz. So for two channels that's around 1.5GMAC/s.

I'm not very familiar with FIR but to keep the 50 Hz resolution you took as an example, it will need less than 400MMAC/s at 96 kHz sampling rate which is the internal sampling rate of the DLCP or am I wrong ?

An alternative that greatly reduces processing combines a tapped delay line and several IIR filters to approximate the inverse all-pass function. This is much more economical but it means a lot of work for the poor bloke who has to write the design software. This poor bloke is called Jan-Willem so go easy on him

this would be a very interesting way, could you develop it a little more if you don't mind.

Are you using the TAS3108 for the DSP ?
 
The rule of thumb linking length of a FIR filter to frequency resolution is just that: a rule of thumb. It says "you can do something meaningful down to this frequency". What or how meaningful depends on the situation. Unless the thing you're trying to do has a response no longer than 1 cycle, the rule of thumb is incorrect.

In the particular case of correcting an LR4 sum what you want to make is the impulse response of a 2nd order all-pass filter with Q=0.7, reversed in time. The impulse response of course is not finite. It keeps going. So you need to decide how far it should decay before you stop it (or window it down). The simple rule-of-thumb would have you stop it after a single cycle, which is much too soon. It would produce unacceptable ripple in the magnitude response, and it would not be very accurate at all.

How far you let the response decay depends on how accurate you want the result to be. Since the filter will affect the whole audio band, you cannot afford much ripple (i.e. echo). The filter will be longer than expected. Perhaps a bit shorter than I proposed (that was based on letting the impulse response die down about 150dB), but the order of magnitude is right.

The DSP is indeed a TAS3108, which has about 135MMAC/s to play with.
 
Last edited:
You most definitely do not need FIR filters to iron out phase errors of the drivers since correcting the corresponding magnitude response errors does the phase bit for free.
Assuming the drivers are operated well within their passband, I agree. In practice, it's been my experience that's a luxury since the phase response to durable frequency shifts---crossovers, shelving filters, rolloffs due to Fs and drivers becoming acoustically large---has a greater spectral extent than the magnitude response. For example, I've worked quite a bit with some tweeters which have a roughly fourth order roll off below an Fs of 650Hz. The phase variation from this remains significant for crossovers up to 2.5KHz or so and requires some form of delay equalization to get the tweeter and mid to sum properly. The design happens to be one where electronic delay is preferable to a physical offset. (The tweeter does have four min phase IIR biquad patches and, as those are in band, they flatten out the phase response as well without need for delay EQ.)

I was surprised at how vanishingly small the audible effect was. It took me quite a bit of time to find music that would actually make it detectable at all. Linear phase is hugely overrated.
I'm not aware of any fully rigorous data on the audibility of linear phase, but some small sample size blind studies have been done. Their finding was roughly a third of folks prefer linear phase, roughly a third of folks prefer min phase, and roughly a third of folks don't hear a difference. Any sense if you happen to fall into the latter category? Also, I get the impression the linear phase design you did was LR2. I haven't worked with LR2 much but my experience with LR4 and LR6 is the phase rotation of min phase LR4 responses is less audible than that of LR6. Presumably due to LR4's lower group delay. It's possible the group delay variation of LR2 isn't big enough to be particularly audible.

Don't have time right now but if need be, I can do a couple quick experiments to test LR2 audibility that should be reasonably robust. Personally, I've scored 100% discrimination between linear phase and min phase LR4 and LR6 in my attempts to approximate blind ABX testing---to my ears the difference is obvious, linear phase clearly preferable, and the results robust enough the correlation good enough I'm willing to rule out placebo effects.

This is much more economical but it means a lot of work for the poor bloke who has to write the design software. This poor bloke is called Jan-Willem so go easy on him ;)
Tell me about it; I'm working on solving the same problem and have cases where delay equalization down to 20Hz is desirable---on a stream based DSP so windowed FFTs and time reversed IIR aren't options. Will toss Jan-Willem an email and see if we can strike up a collaboration.