So here's where I want to talk about some digital filtering. Specifically, I want to hear some thoughts/get a discussion going on these guys: <link>.
Right now, I'm working on a preamp design based on the PCM2707 and the PCM1792a. I want to stick the QF's right in between those guys to achieve some crossing-over in my preamp. The whole thing is microcontrolled (PIC18LF14K50).
What does everyone think? I've been playing around with the design software (FREE! <link>) and I've got a few questions to get things going:
How good does the filtering need to be at the preamp stage for some premium multi-amping? To put it another way, is more aggressive filtering always an improvement? For instance, how’s it gonna sound if I give my my mids exactly (or as near as makes no difference) 1000-6000 Hz--a theoretical example…assuming the drivers deliver their most linear performance in this range? If it is better to have some overlap, is there an ideal amount (characteristics of the roll-off)? This is something I’m excited to tinker with once (if) it’s built, but maybe someone out there has opinions/data to target the tinkering.
Certainly, anything you have to add is something I'm interested in. Especially look at that software and let me know what you guys think!
Right now, I'm working on a preamp design based on the PCM2707 and the PCM1792a. I want to stick the QF's right in between those guys to achieve some crossing-over in my preamp. The whole thing is microcontrolled (PIC18LF14K50).
What does everyone think? I've been playing around with the design software (FREE! <link>) and I've got a few questions to get things going:
How good does the filtering need to be at the preamp stage for some premium multi-amping? To put it another way, is more aggressive filtering always an improvement? For instance, how’s it gonna sound if I give my my mids exactly (or as near as makes no difference) 1000-6000 Hz--a theoretical example…assuming the drivers deliver their most linear performance in this range? If it is better to have some overlap, is there an ideal amount (characteristics of the roll-off)? This is something I’m excited to tinker with once (if) it’s built, but maybe someone out there has opinions/data to target the tinkering.
Certainly, anything you have to add is something I'm interested in. Especially look at that software and let me know what you guys think!
With steeper filter slope you'll (ussually) have less cancellation problems off-axis. But steeper filters in analog or recursive digital filters will result in more group delay at the crossover.
With FIR filters you can implement linear phase filters, which have zero group delay. However, some studies have pointed out that using long linear phase FIRs will result in audible pre-echo's and ringing due to the symmetry of the impulse response.
So, there is no ultimate answer here, totaly depends the type of crossovers you need, the cutoff frequencies and the layout of the drivers.
With FIR filters you can implement linear phase filters, which have zero group delay. However, some studies have pointed out that using long linear phase FIRs will result in audible pre-echo's and ringing due to the symmetry of the impulse response.
So, there is no ultimate answer here, totaly depends the type of crossovers you need, the cutoff frequencies and the layout of the drivers.
Hey, transphere. Thanks for the input! This is exactly the type of stuff I was wondering about. You think you could point me to some reading material?
Also, would a concrete example allow you to give me a stronger opinion? Let's say I want to use the following:
Eminence alpha 8a 100-800Hz
Eminence alpha 6a 800Hz-2kHz
HiVi RT1.3WE 2kHz-->
My instinct is to filter as sharply as possible at the crossover frequencies with the sharpness limited only by pristine flatness in the passband. Is there an advantage to overlap the crossover frequencies slightly with a softer rolloff?
Well, you can use either regular type (Linkwitz-Riley, Butterworth, Bessel) crossovers, and with these you're probably limited to a maximum 6th order (36dB/oct) or so before group delay would become objectionable. The other option is to use linear-phase brickwall filters, where you have zero group delay but also possibly audible pre-echo's off-axis. You could design linear phase filters with softer rolloffs, but you would trade in the pre-echo's for off-axis cancellation problems. Here you can find a study on the audibility of crossover artefacts for both Linwitz-Riley and linear phase brickwall crossovers.
One practical factor to consider in this case is if the 512 FIR taps of the chip will be enough for true brickwall performance at the crossover frequencies. If not, you'll have trouble finding a complementary filter pair, which is not that trivial.
To keep things simple I would first investigate if mapping a normal Linkwitz-Riley crossover (which as an infiniite impulse response) onto a finite impulse response filter gives you enough resolution. Since you'll be truncating the impulse response to fit it in the 512 taps, both phase and magnitude response of the filters will deviate from the prototype Linkwitz-Riley response. If 512 taps is not enough you'll notice having a non-flat summation and a distorted polar pattern. In this case you would probably want to switch to linear phase filters.
Since you're using this chip, you can make an implementation of both, try both programs and pick the one you like the best.
One practical factor to consider in this case is if the 512 FIR taps of the chip will be enough for true brickwall performance at the crossover frequencies. If not, you'll have trouble finding a complementary filter pair, which is not that trivial.
To keep things simple I would first investigate if mapping a normal Linkwitz-Riley crossover (which as an infiniite impulse response) onto a finite impulse response filter gives you enough resolution. Since you'll be truncating the impulse response to fit it in the 512 taps, both phase and magnitude response of the filters will deviate from the prototype Linkwitz-Riley response. If 512 taps is not enough you'll notice having a non-flat summation and a distorted polar pattern. In this case you would probably want to switch to linear phase filters.
Since you're using this chip, you can make an implementation of both, try both programs and pick the one you like the best.
- Status
- Not open for further replies.