TOPIC:

This is the story of upgrading from an analog active crossover to a miniDSP Flex 8.

I built a set of 3-way speakers in 1994 with a 6th-order woofer tuning as described by A.N. Thiele in his article on ported-box speaker tuning.  This type of tuning requires a 2nd-order active high pass filter on the signal sent to the woofer.  I had already decided to use an active crossover network, so this fit into my plan easily.  I generated my own design tables and selected a tuning that would reduce the box size without sacrificing bass depth.  The source (most recently a PiCorePlayer with a Matrix USB->SPDIF converter feeding an MSB Nelson Link III DAC) fed into my active crossover unit, which featured a source selector switch and volume pot.  I hand-wired the active filter circuit on a pad-per-hole 0.1" grid prototype board with a ground plane on one side.  Back then you couldn't have a custom board manufactured for you in a week for $50.  I used a Sallen & Key filter design, which provides a 2nd order filter for each op-amp, to implement a 4th-order Linkwitz-Riley network.  The design passes the tweeter signal through 2 op-amps, the midrange signal through 4 op-amps, and the woofer signal through 3 op-amps.  These started out with Burr Brown OPA2604 audio op-amps, and because they were socketed I was able to upgrade as technology progressed to OPA2134 and last year to OPA1642. 

The crossover board:


The crossover fed its output into a Classe CAV-75 six-channel power amp, which was connected directly to the individual speaker drivers:

Woofers: Audio Concepts AC10
Midrange: Vifa P13WH-00-08 made for Madisound
Tweeter: Originally Vifa 25mm aluminum dome, then upgraded to Focal inverted titanium dome.  When the foam on the Focals failed, I replaced them with SB Acoustics SB29RDC-C000-4.

 

In the early 2000s, I was excited to learn that Alpine's car stereos used digital crossover filters.  I bought one for my car, and started dreaming about designing my own digital crossover using the Burr Brown filter chips.  Maybe even integrating D/A converters.  I never did, but last month a friend suggested I check out miniDSP and I saw that now, I could just buy the device I had been dreaming of building.  So I ordered a Flex 8 with Dirac Live.

Setup was a piece of cake.  The Linkwitz-Riley network was easy to set up using the "basic" crossover menu.  To implement the woofer auxiliary filter, I initially used the advanced mode crossover setting.  I used the spreadsheet linked from miniDSP's application notes to calculate the biquad coefficients for the auxiliary filter from the Q and fc from my original design notes (well done, 30-years-ago me, for taking such good notes!).  Because I switched the woofer crossover to advanced mode to do this, I also had to use the spreadsheet to calculate biquad coefficients for the Linkwitz-Riley low-pass filter.  Later on, I figured out that I could use one of the parametric EQ biquads to implement the auxiliary woofer tuning filter, so that the woofer crossover could stay in basic mode! I set the relative gains of the drivers based on the manufacturers' sensitivity specs.

Setting aside the USB->SPDIF converter, Link III DAC, and my crossover box, I plugged the PiCorePlayer into the USB input of the Flex 8 and hooked the Flex 8 analog outputs 1-6 to the power amplifier channels.  It required some tinkering with the PiCorePlayer device driver settings, but once it knew how to talk to the Flex 8 I was rewarded with a working crossover with digital volume control (with a REMOTE!).  All the right kinds of sound came from all the drivers.  It was very good sound, too.  I thought it already sounded better than the system sounded with the analog crossover.  (Note: I had auditioned a Topping D90LE DAC against my MSB Nelson Link III last year and concluded that in A/B testing I could not hear the difference between the near-perfect-measuring D90LE and the Link III.  Nice output DACs, miniDSP!)

After a few days, I downloaded REW software and followed the Loudspeaker Measurements application note to measure and correct the individual drivers.  I found that the gating technique was necessary to obtain a meaningful response curve.  The 3.4ms reflection delay meant that I only got frequency measurements valid down to around 200 Hz, so I was only able to correct the tweeter and midrange.  I found that the sensitivity specs from the manufacturers were dead-on.  With the gain settings I had made based on those specs, both tweeter and midrange were within 0.1dB of each other along the flattest parts of their response curves.  I could have corrected the woofer response from 200 Hz up, but since the crossover frequency was 350 Hz, I didn't bother.  (Thinking about this now, I think it could still be beneficial to equalize the upper range, and I could verify the gain matching). 

Next, I measured the relative time delays of the drivers.  The first thing I noticed was that the pulse at 1ms was upside-down.  My left midrange was connected with the terminals reversed!  Everything else was in phase, but my left midrange has been out of phase for at least 9 years.  Derp.  I guess that explains the indistinct stereo imaging I'd noticed.  It's much improved now, with the polarity correct!  The delays were pretty minor -- about 0.08ms tweeter to mid, and another 0.08ms mid to woofer.

With all that taken care of, the system sounds glorious!
 

A few days later, I had a chance to run the Dirac Live measurements.  I did the "standard" measurement set (center, 8 cube vertices, 2 measurements farther out to each side).  I did not have anyone seated in the center position -- I hid on the floor behind the sofa for each measurement.  The overall response roughly matched the target curve that was drawn, with a few minor wiggles.  I had it generate the filters based on the default target. 

I was not impressed by the sound with the Dirac Live filter. It sounds fine.  It's very neutral sounding, but it sounds less dynamic and lively.  Without the filter, there's more sense of air and the lower register of human voices sounds richer and more lifelike.  Low percussion has more impact.  It's not a subtle thing, switching back and forth.

Knowing what I know now, if I could re-spec my purchase I would still have gotten the UMIK-2 calibration microphone but I would not have paid the premium for a Dirac Live license. 

I've put the Flex 8 into non-Dirac mode (96k, user access to the FIR filters) and transferred the settings over by hand (recalculating the advanced mode PEQ biquad for the woofer auxiliary filter because the coefficients depend on the sample rate).  This is where I'm going to stay, at least until I can get the speakers outside and onto a raised platform so I can extend the frequency window of my measurements lower to get better corrections for the mids and woofers.