NO! Consider it mono and start over! Read whole page to see why. 🙂
This graphic equaliser was made for the bass sections of an active three-way speaker system and was built to meet the following goals:
- To provide independent left and right channels with plenty of bands spread across the pass band of the woofers.
- To work for music played by the PC as well as other playback devices.
- To be installed in-line after the active crossover’s low pass outputs and in front of the woofer power amplifiers.¹
- To be completely transparent when powered off or switched to bypass.
- To be set once to impulse response measurements taken at the sweet spot, then sealed.
- To have no exposed controls other than the power and bypass switches.
- To be built inexpensively – using left-over and recycled parts wherever possible (in case of abject failure).
- To be in a chassis of its own, so if it didn’t work as hoped it could be “thrown on the tip“. 😆
1. This was a mistake. The EQ must be applied ahead of the active crossover! Otherwise adjustment of the upper bands can upset midrange crossover alignment – both amplitude and phase.
Toole says that stereo bass is “much ado about even less”, but I still think it’s important to achieve an in-room amplitude response in which the left and right channels sum acoustically to something at least half decent.
The system had been sounding pretty decent since tweaking the respective pass band levels, so I didn’t want to ruin it. Nonetheless unlike the downstairs system with “hard-wired” eight-band stereo graphic EQ, the in-room bass response didn’t measure anywhere near flat at the sweet spot and significant variations between the channels were evident. Here is what I mean:
Red is the right speaker and blue is the left as measured individually at the sweet spot some 4m from the speakers. These were taken with a 500ms window, so it’s not the direct sound (which arrives in about 11ms), but it gives an idea. I suppose the biggest problems with the raw responses are those 40 and 63Hz “booms” or modes excited by both channels, so when played loudly, the overall impression is dominated somewhat by them.
The room is quite asymmetric with bay windows close to the left speaker and a staircase void and open space to the right.
It turns out that these angled walls are actually good and do provide a kind of damping effect.
But something had to be done. Commercial “room treatments” including so-called “bass traps” and “diffusers” are quite ridiculous in a domestic environment and don’t work any more effectively than ordinary household items such as fabric covered lounges, curtains, carpets and general decor anyway. Opening the windows would be far more effective than any commercial “bass trap” available, but that’s a little impractical. Some interesting active traps (with speakers driven in anti-phase) were developed in the 80’s, but never took off. EQ can work well for a specific listening position, but compromises the overall power response of the speakers in the room. It could ruin the sound everywhere other than at the sweet spot, so I decided that if used, it must be capable of being bypassed for group listening.
My music library is played using JRiver Media Centre which has IIR parametric EQ available in its “DSP Studio”. It’s quite sophisticated and allows you to enter as many filters as you want and provides separate capability for each channel. As a preliminary exercise I re-measured the bass response at the sweet spot for each channel and used the measurement software to generate the correction filters. For the left channel it looked like this:
The dark blue trace is the measurement. The faint dotted line is the prediction and the pretty coloured areas are the individual filter corrections. The blue filled-in area represents the sum of the corrections. If you invert them the summed area approximately fills in under the dark blue measurement trace:
Again – this was taken at the 500ms “window”, so aren’t representstive of the direct sound.
Note: this separate left and right speaker measurement technique turned out to be stupid, but read on anyway.
The filters are represented by a table of numbers (Fc, Q and boost/cut in dB) for each filter – in this case 7 filters (the right channel had 8). The Qs ranged from around 2 to 20! These were typed into JRiver’s DSP Studio Parametric EQ tables for “corrected” playback.
The actual in-room “corrected” response is unverified. I.e. I don’t know how to run fresh measurement sweeps through JRiver’s loop-back feature and I’m not really interested since this is just a trial, so I just assume that it does things as promised. Initial impressions were that the improvement was spectacular. The removal of all that “boom” seemed to open up the midrange/treble imaging for an epic improvement in the sound stage. It seemed like it removed all the fun too though, but no – it took a little getting used to, but the fidelity was pretty extreme without the boom. * However – see update below.
That’s all well and good, but what about when friends come over with CDs? An analogue stereo equaliser could still be switched in for those occasions.
The analogue approach uses a pair of old ESP P84 PCBs (⅓ octave band graphic equaliser with 8 fixed centre frequencies and fixed Qs). Rather than just building it however, it could be simulated using the software.
Here is how it would look for the right channel with the table of values restricted by the fixed Qs and band centres of a P84 with typical resistor and capacitor values:
The inversion doesn’t fill the measurement trace as well, nor match the flat prediction of the parametric EQ tables of course, but it’s not too bad and I doubt I’d hear a difference. Notice it cannot correct the 90Hz suck-out.
Although it might be assumed that if each channel is set individually to a fairly flat response, this would not guarantee a flat response with both channels driven simultaneously with music having substantially monaural bass content. There is conflicting information on the ‘net about multiple subwoofer placement etc. There will be some left/right speaker reinforcement/cancellation at different frequencies and this could really complicate things. Once each channel was set individually, more sweeps through a Y-splitter will be needed to see how the summed responses pan out. My guess is that some fine tuning might flatten it further, but would that be “correct”, or should it be left at the stereo setting? I really don’t know yet.
The obsolete P84 “Version 1” PCBs were “thrown in” when I purchased a couple of the upgraded double-sided PCBs from ESP some years ago for the Jamo oriels. These old ones had been sitting around. I also had a transformer and a P05A power supply in something that was no longer needed.
There were also heaps of unused OPA2134s, so all I needed was a chassis, a bypass switch, a relay, components to populate the PCBs as well as LEDs, connectors, stand-offs, a fuse holder and wires etc. The whole thing took less than a week to assemble once all the parts were scrounged.
The chassis had to match this one because it’s to sit on it, so I ordered the same one from VT4C in Hong Kong. It arrived in a couple of days.
I told a little fib about the thing needing a bypass switch. The standard chassis has a big recess to take a volume control knob and that needed to be filled with something – so a rotary bypass switch it was.
Here are some photos.
Just 4 sockets – 2 in and 2 out. When powered down, the inputs are coupled directly to the outputs via the NC relay terminals. A solid copper wire grounding bus around the RCAs:
Those Teflon coaxes each have their shields connected at one end only. The signal ground/PSU zero Volt line is isolated from the chassis/mains earth by the RCA socket plastic washers.
The rotary switch has 4-poles, so I connected them in parallel and soldered the LED and relay coil voltage drop resistors directly to some of its otherwise unused terminals:
The equaliser PCBs were a bit of a pain being early non-solder masked single-sided boards with a very fine and complex solder side and a few manufacturing/design faults, but these were sorted out and they work perfectly with the scope showing nice clean traces with all bands behaving as hoped:
Time to calibrate
My brother just purchased an Earthworks DM23 microphone with a calibration file specific to it, so it was time to calibrate something with a little more confidence. To date (30 September 2013) all response graphs shown on this site (including the above) were made using a cheap Behringer microphone and a generic (poor) calibration file.
The approach was to repeat the impulse response measurements with the Earthworks microphone between adjustments of each of the eight cut/boost bands (25, 31, 40, 50, 63, 80, 100 and 125Hz) until the measured response was as flat as reasonably possible for each channel, then screw down the lid. That is, the actual measured results verify the performance during the calibration process. For each channel, I started by setting the bands requiring the largest cuts, then tweaking the less problematic bands. Of course there are a couple of places where deep/sharp notches are present. These regions of “excess group delay” were ignored as boosting can’t fill them.
Here is a preliminary result for the left channel:
Blue is EQ bypass mode and red is after numerous adjustments (toward flat).
And for the right:
The room response of the left channel had a better overall EQ, but the right channel has deeper extension. These seem to match the predictions reasonably well and confirm that the 95 Hz suck-out in the right is not correctable. Note that these measurements were made with a fairly wide IR window or “gate” and therefore do not ignore the many room reflections (reinforcements/cancellations). The graphs do not represent the direct sound. Sitting at approximately 4m from the speaker front baffle the direct sound arrives in about 11ms. So is it right?
I’ll report some subjective listening tests later…
During music playback, the transition between boom and no boom was unquestionable. The EQ was probably about as effective subjectively as the parametric “DSP Studio” version. Not sure really.
30 October 2013 – Room for improvement?
Well it’s wasn’t over yet. And something seemed wrong. There was a certain “thinness” to some music and I still had not looked at the left/right summed acoustic response and this is what you hear after all.
So I decided to run a sweep signal through a Y-splitter to measure both channels driven simultaneously and here it is (in all its horror):
A disaster area! The teal trace is in bypass mode and the dark purple trace is with summed stereo EQ! I dialled in a bit of a “house curve” in which the SPL increases slightly to the left, but what the heck is that at 72 Hz?
Think, think, think….
OK, I figured that I needed to regard each stereo band as one. E.g. If the left 63Hz pot, it turned 10 degrees, then the right 63Hz pot needs to do the same. More to come.
Ok, done. And it’s bizarre! It turned out that a setting on one of the Hypex plate amps (powering the subwoofers) was mainly to blame. In earlier fiddling, I had deliberately turned the phase dial on the left sub-woofer plate amp to 90°. This improved the crossover response of the left speaker slightly (removed a slight suck-out at the 150Hz crossover frequency), but since the right plate amp was left at 0° (where its in-room crossover response was fine) there was a left-right interaction that I had not anticipated!
Anyway, I started over by turning the plate amp phase dials to the same positions. Then I raised the bass levels at the active crossover by 1.5dB. Why? Two reasons: The correct/designed crossover point is 150Hz, but measurements had this shifted a little to the left (down). Raising the level shifted it to the right where it belongs. Also for the sake of actually seeing the very bottom end response as “tuned” by the Linkwitz Transform and with a view to keeping that nice “house curve”, it was better raised a bit.
OK. Here are the new “both channels driven” curves:
Looks good! Red is with no EQ. Blue is the response of both channels driven simultaneously after many EQ band adjustments made equally to both channels. At the end, each EQ band was confirmed to match electrically using the oscilloscope.
But look at the individual left and right responses after doing that EQ (channels driven independently):
Purple is the left channel and orange is the right.
Far out! 😮
In the modal region measure with both channels driven simultaneously then EQ the same for both channels. It’s all about coherence – do the same to every sub.
Intuition: Don’t mix stereo down to mono for playback though. The deep bass content of most stereo recordings is coherent left-to-right, so it’s as good as mono anyway. But some music and especially some electronic music and some specially mixed “effects” recordings (like some on the Telarc label) have discrete deep bass content left and right. Mixing would wreck these IMO.
Additional Photos June 2020
The equaliser was just tweaked to shift the 80Hz centred band down to 73Hz with a Q of 5 to address a particular room mode, so I took some better photos. There is also a second relay for muting the output on powering off. The little perforated board just applies an un-muting delay to the additional relay at the back panel: