Hi,
I've had a Najda DSP crossover sitting in on my shelf for years. Finally hooked it up yesterday, lol.
I'd like to know if there is a way to use longer FIR filters? It limits me to 1024 taps. Since I would like to do some phase adjustment in the low frequency range it looks like 2000-3000 taps is needed.
Also, why the limit?
Thanks!
I've had a Najda DSP crossover sitting in on my shelf for years. Finally hooked it up yesterday, lol.
I'd like to know if there is a way to use longer FIR filters? It limits me to 1024 taps. Since I would like to do some phase adjustment in the low frequency range it looks like 2000-3000 taps is needed.
Also, why the limit?
Thanks!
Longer filters need not only more memory (not really an issue nowadays) but also more processing power (probably the bottleneck in this case).
Without additional DSP horsepower its quite possible to implement longer filters in the bass by means of decimation. But that means more complex software.
Without additional DSP horsepower its quite possible to implement longer filters in the bass by means of decimation. But that means more complex software.
I plan to use one Najda per speaker, so processing power should be okay.
Can you please explain more about your second point? How can I do it? I planned to create my FIR filters in Re-phase and load in to the Najda as a FIR block however it just says 'filer too long'.
Can you please explain more about your second point? How can I do it? I planned to create my FIR filters in Re-phase and load in to the Najda as a FIR block however it just says 'filer too long'.
If you're not a DSP programmer you'll not be able to implement decimation - I'm not a Najda user myself but I would bet its not supported in the standard software. To get a longer filter by decimation you'd need to delve down to the bare metal.
Decimation means reduction of the sample rate - in essence you want to implement a crossover akin to a subwoofer crossover by software. The high pass section just needs delay, the low pass section you reduce the sample rate of and then, at the newer low sample rate you implement your bass filter. (Given the sample rate's so much lower you have more horsepower available for more taps.) After that, you upsample and re-combine with the high passed signal.
Decimation means reduction of the sample rate - in essence you want to implement a crossover akin to a subwoofer crossover by software. The high pass section just needs delay, the low pass section you reduce the sample rate of and then, at the newer low sample rate you implement your bass filter. (Given the sample rate's so much lower you have more horsepower available for more taps.) After that, you upsample and re-combine with the high passed signal.
I see, thanks for that!
It sounds like you know about DSP filtering so I wonder if you can answer a different question?
On a smaller DSP board I did not have enough processing power for the length of filters I really wanted to use. Of course, the high pass filter does not need to be as long as the low pass, but if I used a shorter filter for the high pass, the linear phase xover does not match up.
Is there a method to use different filter lengths for high pass and low pass but make them match? Is it as simple as a delay by the difference in samples?
It sounds like you know about DSP filtering so I wonder if you can answer a different question?
On a smaller DSP board I did not have enough processing power for the length of filters I really wanted to use. Of course, the high pass filter does not need to be as long as the low pass, but if I used a shorter filter for the high pass, the linear phase xover does not match up.
Is there a method to use different filter lengths for high pass and low pass but make them match? Is it as simple as a delay by the difference in samples?
I would guess (without knowing the details of the software you were running) that yes, it is as simple as delaying the high pass to match the delay of the low pass. With linear phase filters the delay of every filter is half the number of taps (because linear phase are always symmetrical).
For example if your low pass was 128 taps and your high pass 32 taps then the difference in delay is (64-16) = 48 taps.
For example if your low pass was 128 taps and your high pass 32 taps then the difference in delay is (64-16) = 48 taps.
The fact it's limited to 1024 which is a power of two suggests the limit is in the coefficient memory, not the clock speed. The hardware may be a DSP processor with limited RAM or an FPGA with RAM blocks.
From the limited technical info I've been able to find, its a dual core 250MHz DSP but I haven't found which. 1024 words of coefficient memory would only be 4kbytes if the coeffs were 32bits (or single precision floats). That would seem to me a pretty severely limited DSP nowadays but I'll continue to track down the chip that's being used.
Looks to be based on one of the old Motorola 24bit DSPs : DSP56725|24-bit Dual-Core Symphony(R) DSP | NXP Semiconductors