MiniDSP crossover development

[johnk...]

Re: MiniDSP crossover development

Comparison of analog active to the miniDSP 2x8 is something I might demonstrate at the next DIY New England. I have my NaO II RS designed with options for a fully active analog crossover and using the miniDSP 2x8 digital. The transfer functions of both are identical in terms of the specification (Q, corner or notch frequency, gain or cut). They measure within a fraction of a dB of each other. I place both between the preamp and power amps. Both sound very good but I find the analog to be a little better though without direct comparison I probably could not tell you which crossover was in the system.

I used Charlies spread sheet to generate the byquad coefficients.

[frangus]

Re: MiniDSP crossover development

Hi Charlie,

Did you develop this any further? It's exactly the kind of software i'm looking for..

[charlielaub]

Re: MiniDSP crossover development

Hi Charlie,

Looks good. How's it sounding?

Dave

Hi Dave,

Although I only have one speaker put together, it sounds pretty good. There are a few things I want to tweak: I need to add some damping material to the walls, but the cabinet is nicely bolted together and seems to not have any bad internal resonances. I could cut a slightly longer port tube to reduce the box tuning frequency a bit. I will probably increase the baffle step about 1 dB and play with the frequency a little - this should influence the mid to upper bass balance. Unfortunately, I could not measure to a low enough frequency to capture the full baffle step, so I had to model it. Once that is dialed in, I can remove the baffle, remove the drivers, and spray on some paint to both baffles.

I've always wanted to build a 2-way with an 8" woofer. I had a pair of Snell K IIs back in the '90s and I really liked them, so this is my effort to capture the memory of my speakers.

-Charlie

[DDF]

Re: MiniDSP crossover development

Hi Charlie,

Looks good. How's it sounding?

Dave

[charlielaub]

more speaker crossover developement to report...

I'm continuing to develop and refine my Excel tools. While I am at it, I am also trying to build some speakers! So here is an example of my latest efforts.

For this speaker, I am using the IKEA "baffleXchange" concept that I developed a while back. The cabinet is about 50 liters total volume. I am using a Peerless 850137 8" CSX woofer and the Dayton RS28F tweeter. The cabinet is ported with Fb=34Hz.

I did some modeling of the driver positions using Jeff Bagby's diffraction modeling tool, and I found that this offset tweeter position help to boost the tweeter response around 1k - 1.5k Hz, and results in the off axis response being smooth and independent of angle. The best setup for listening is with the tweeter offset towards the "inside" of the stereo pair.

I played around with various versions of the crossover, including delay, slope, etc. I ended up using a 4th order/4th order crossover plus a 1k Hz notch on the tweeter. There is also a first order shelving filter to lift the bass response back up to the level of the mid and high frequencies, and a second order high-pass filter on the woofer to reduce cone excursion below tuning. The notch steepens the rolloff below 2k Hz considerably. I could definitely tell the difference during auditioning at higher SPLs, with the notch sounding much less strained (if at all), while a 4th order 1.5k Hz crossover sounded a bit strained. Another benefit of the notch is that is helps with the phase alignment with the woofer so that, after juggling the crossover points a bit, I didn't need to use delay on the tweeter. This will simplify the final analog crossover implementation. The crossover point ended up being at about 1775 Hz, which should be fine for the RS tweeter and is still low enough that the woofer is not yet beaming too much.

In the frequency response plot below, the white line is the response measured with ARTA on axis with the tweeter at 1m, the blue line is the response predicted by my tools, and the other lines are as shown in the figure legend.

-Charlie

[neildavis]

Re: MiniDSP crossover development

It's only like PCD in the fact that the input is FRD files and you can view the driver, filter, and system responses while you are developing the crossover. The actual look and feel is completely different, however.
-Charlie

That's what I was doing about 5 years ago, but nobody seemed too interested :rolleyes:. Take a look at the old PE post at the link below. It points to an Excel spreadsheet that reads FRD files, allows you to specify filter configurations for up to 7 biquads in a TAS3004 chip, and then it generates the biquad coefficients from the filter specs. Then, it converts the values to HEX strings that can be sent directly to the EEPROM on a board with a TAS3004 audio processor chip. Hit the reset button on the spreadsheet and it will load the EEPROM into the TAS3004, so you can change the DSP programming directly from the spreadsheet, in "real-time".

I've got a later version that programs the chip via a USB connection, but you need to have a microcontroller on the board to interact with the program (same as what miniDSP does). I've got code for directly programming the STA308a DSP and other ST/Apogee chips, but I don't yet have code for the Analog Devices chip used on the minDSP. That, plus code to directly control the DCX2496, are on my to-do list.

Here is the link (note the date):

http://techtalk.parts-express.com/showthread.php?p=1401986

This code is very rough and not well documented, but you are welcome to use it. The spreadsheet is not protected in any way. This code went through a translation to .NET and got cleaned up and well documented for my active loudspeaker design suite, so it is probably not too useful as-is.

BTW, at that time there were a number of people having some problems with PCD importing FRD files, so I made the file import code "bullet-proof". It doesn't care if there are blank lines, extra lines or extra characters, and the delimiters can be just about whitespace character. It also has a good interpolation routine for resampling the data to 500 sample points. This version is very hard to read, but the .NET version is nice and modular and easy to follow.

[evilskillit]

Re: MiniDSP crossover development

That would be amazing!

Hi Dave,

Yes, that's the general idea. The input is FRD files. The user describes the filters by inputting the frequency, Q, and gain of to be applied to each driver. The output is given in terms of the transfer function coefficients for both the analog (s) and digital (z^-1) domains. The latter is formatted for the MiniDSP so that it can be directly pasted in to the "advanced biquads" dialog box in the MiniDSP interface.

This makes for very rapid updates to the MiniDSP, and one can play around and try out various crossovers very easily.

It's only like PCD in the fact that the input is FRD files and you can view the driver, filter, and system responses while you are developing the crossover. The actual look and feel is completely different, however. I've not used any macros, which I am hoping will make compatibility issues between different versions of Excel a non-problem. It may even run on OpenOffice Calc...

-Charlie

[devinkato]

Re: MiniDSP crossover development

charlie - this is very exciting news. I have a several minidsp units that I've used in various basic scenarios, and it is awesome to hear that there may be "formal/developed" tools that will help me take it to the next level.

[greywarden]

Re: MiniDSP crossover development

I say develop it on Open Office Calc. It would be more advantageous to everyone. Anyone can use it without the need of expensive software.

[charlielaub]

Re: MiniDSP crossover development

Congratulations Charlie.

If you ever make these tools public domain (ie PCD that takes mini-DSP formatted inputs), I'll be an early adopter and buy the mini-DSP boards.

This is much more attractive to me that buying a new measurement/design suite with a long steep learning curve.

Dave

Note: I edited this post to provide more details on the design spreadsheets. -Charlie

Hi Dave,

Yes, that's the general idea. The user supplied measured data as FRD files. There are two different types of spreadsheets. In a "Driver Response" spreadsheet, the user pastes in the driver's FRD file and describes the frequency, Q, and gain of the filters that will be applied to each driver. Plots of the raw and filtered driver response and phase, as well as the filter magnitude and phase are provided and the transfer function coefficients are listed for both the analog (s) and digital (z^-1) domains. The latter is formatted for the MiniDSP so that it can be directly pasted in to the "advanced biquads" dialog box in the MiniDSP interface. In a separate spreadsheet, the "System Response" is calculated as a sum of all the driver responses. The user needs to paste (as a link) two columns that represent the response of the driver+filters from each driver response spreadsheet into the system response spreadsheet. To model a system, there is one driver response spreadsheet per driver, and one system response worksheet, and all of these open should be open at the same time. I set Excel's calculation to "manual" and then trigger a global recalc once I have finished making my changes. Recalc time on my 3 year old computer is about 4-5 seconds. Once I am satisfied, I copy out the MiniDSP biquad coefficients from each driver response spreadsheet and paste them in to the MiniDSP interface. I can then immediately listen to the speakers.

This makes for very rapid updates to the MiniDSP, and one can play around and try out various crossovers very easily.

It's only like PCD in the fact that the input is FRD files and you can view the driver, filter, and system responses while you are developing the crossover. The actual look and feel is completely different, however. I've not used any macros, which I am hoping will make compatibility issues between different versions of Excel a non-problem. It may even run on OpenOffice Calc...

-Charlie

[DDF]

Re: MiniDSP crossover development

Congratulations Charlie.

If you ever make these tools public domain (ie PCD that takes mini-DSP formatted inputs), I'll be an early adopter and buy the mini-DSP boards.

This is much more attractive to me that buying a new measurement/design suite with a long steep learning curve.

Dave

[charlielaub]

Re: MiniDSP crossover development

I agree with you on miniDSp's software, The other thing I don't like about their software is that they don't uses the usual analog filter definition for Q in the peak/notch filters. They did make a special version for the of the advanced 2 way plugin for me though so I could directly transfer the settings for the NaO Note to their boards. But even so, I find I have to dick around with fine tuning to get the digital transfer functions to match my analog ones to within acceptable error (less that 1/2 dB).

You could program the peak/notch as yet another biquad and not use their built-in EQ functionality... you get 8 biquads per channel as part of the crossover section, but there are an additional 6 per channel in the EQ section for each driver and another 6 on the input EQ (in the plug-in that I use), which is plenty even for a driver like a midrange where you might have 2-3 LP filters and 2-3 HP filters, leaving you with 8 biquads for EQ bands, all pass, etc., etc. plus the 6 on the input. It's a limitation that I can easily live with, given the price.

-Charlie

[benchtester]

Re: MiniDSP crossover development

I am watching with much interest. I am particularly interested in everyones' opinion of the sound quality of analog active crossovers (esp. small custom ones like Charlie's) compared to digital crossovers (MiniDSP, Ultimate Equalizer, and DCX2496).

[johnk...]

Re: MiniDSP crossover development

I agree with you on miniDSp's software, The other thing I don't like about their software is that they don't uses the usual analog filter definition for Q in the peak/notch filters. They did make a special version for the of the advanced 2 way plugin for me though so I could directly transfer the settings for the NaO Note to their boards. But even so, I find I have to dick around with fine tuning to get the digital transfer functions to match my analog ones to within acceptable error (less that 1/2 dB).

[charlielaub]

Re: MiniDSP crossover development

I think software that optimizes cascaded biquad filter functions and generates the biquad coefficients for digital implimentation is a good idea. However, I'm not sure why you woudl want to have boards made for you. I think you will find it very hard to compete with miniDSP. But if you can develop a better product for the same or less money I'm all for it.

Hi John,

At one time I thought that it would be possible to sell my active crossover boards, but that has turned out to be a pipe dream at best and at this point I have purely DIY aspirations. I designed the boards and had a run of them made (I posted about it in the past year or so) but I couldn't scare up interest anywhere. I really made them for my own use anyway - I have always wanted to build active speakers, with amps and active crossovers incorporated in to the cabinet. As part of the design process, I didn't have a good way to "try out" a crossover design before going through the process of building up an analog board (I don't have SoundEasy or other software for that purpose). I prefer to create my own design tools so that I learn about theory and design along the way, from the ground up. Although I've found that the MiniDSPs are awesome tools for implementing filters, MiniDSP (the company) doesn't go very far to make the advanced programming of the biquads very accessible in a format that people can easily make use of IMHO. Even if they did you still need a way to model the entire system of drivers and filters in order to do the crossover design properly. Since I have already invested in the analog boards, that's pretty much the direction that I will keep heading via momentum of effort. I may make my design tools available to other DIYers once they are a little more mature and better documented.

-Charlie