This chassis houses a range of ESP PCBs – namely a P88 stereo pre-amplifier, a pair of P71 Linkwitz Transform circuits, a pair of P09 Linkwitz-Riley crossovers (for 3-way stereo) and a P05b power supply:
The case is a standard unit available from VT4C in Hong Kong, but with the rear plate milled to add 4 extra RCA sockets.
Three series-connected 5V mute relays short the outputs to a ground wire (middle bottom) on powering down:
The toroidal transformer was custom-made and has an integral “flux band” to trap some of the magnetic field which might otherwise impinge upon the audio circuitry:
That provides 15-0-15V AC to the P05 power supply which had to be beefed up a bit to provide sufficient current for all the PCBs.
The DC outputs (and ground) run to a remote terminal block that has a pair of filter caps (at top right in the above photo) from which ±DC and ground leads “star-off” to the various PCBs. They pass underneath the boards whereas the signal cables pass over the boards.
OOPS! There’s a mistake! Those filter caps on the terminal block do nothing. They are 1000μF, but the output impedance of the regulators is too low. A 10Ω series resistor on each rail now forms a decent RC filter that ought to eliminate 50dB + of high frequency noise. See mod:
They drop the rails from ±15V to arould ±13.3V but who cares? The maximum signal output voltage of the crossover is only about 2V so clipping will never be an issue.
The P88 pre-amp is simplicity itself and is just as good as anything on the market, or lauded on the forums. It uses a log taper (Alps) stereo volume control pot:
The stereo pre-amp signals are fed to the separate left and right channel LT (P71) PCBs. These manipulate the stereo signals with a frequency-dependent voltage gain (and attenuation) rising to about 7.3 (about 17.3 dB) near DC to compensate for the sub-woofers’ natural response shown by the blue line below. The Hypex plate amps have a 4th order subsonic filter with F3 at 12Hz to protect the drivers from the low infrasonic components (i.e. it pulls that red line further down at the far left). At 16Hz the voltage gain is around 5.6 (or 15dB). The red line below is the transfer function (expressed in dB) and the green line is the resultant acoustic response in which the flat part of the response curve extends about 20Hz lower than the speaker’s natural roll-off (blue line). The broad peak around 75Hz is completely reversed and careful trimming of component values achieved a perfectly flat green line down to beyond 25Hz with -3dB at 16Hz:
The circuit also swings a natural phase inversion in the speaker response at around 40Hz down to around 16Hz.
The output of each P71 PCB feeds to the first active crossover (P09) PCB:
All of the signal cables are thin Teflon coax with the shield grounded at just one end to prevent ground loops. I chose which ever end was closest to a power supply connection terminal so that any noise picked up by the shields would be shunted straight back to the star grounding point at the capacitor terminal block without passing across the ground plane of any PCB. See above “P88” for example in the previous photo. The coax cables going to the P71 PCBs are grounded at P88 only since the power supply terminal is right there alongside (where one of the yellow wires is GND).
The buffered high pass outputs of the first P09 connect directly to the RCA sockets for connection to the tweeter amplifier. The low pass is unbuffered and tapped off to the second board which has outputs buffered to the RCA sockets for the midwoofer and subwoofer amplifiers:
All opamps are LME49720 in machined pin sockets – now with heat sinks (not present in most of the photos). And despite what the inmates at the diy audio insane asylum might say, these opamps do not oscillate with ceramic bypass caps installed as per the datasheet recommendations and as provided on all ESP PCBs! Indeed the only unity gain set opamps in the whole chassis are the input buffers to the first P09 PCB. Even these do not oscillate, and there are no parasitic-compensating caps across the feedback resistors either.
The LT phase shift is around 11º at the 150Hz sub-to-mid crossover point. Although placing the LT circuits after the LP outputs of the second P09 would remove the LT circuit from the mid and tweeter signals altogether, I preferred to avoid that step in the phase response because it would cause a slight amplitude dip in the summed acoustic output at 150Hz (actually there is very slight phase non-aligment caused by the drivers anyway, but I didn’t want to add to it). With the LT boards “up front”, the same 11º shift is fed to the crossover network so is present at all the outputs and therefore “cancels”. Misalignments clearly visible in the simulated phase transfer function curves are well away from the crossover points and are therefore attenuated by the steep amplitude slopes and do not affect the audible response.
Out of interest, here is the phase magnitude simulation of the speaker system (not showing the effect of the acoustic tweeter inversion, or the tweeter protection capacitor’s effect which I could not simulate with the software, but which would bring the yellow and green lines toward each other, but including the crossovers, LT and subsonic filtering at the plate amps):
Near-perfect alignment of the red woofer and green midrange curves around the 150Hz crossover frequency. I don’t know how the steepness of those curves affects harmonic summation and transient response (“fast bass”), but there is some reading on the subject here.
And to anyone who says that such phase mash produces audibly inferior sound compared to linear phase digital crossovers, I’m not going to argue, but would suggest that everything is a compromise. Here is a the same simulation with all the filters removed for zero line level phase disturbance – just as it would be with linear phase digital crossovers (i.e. the speaker drivers in their enclosures do this by themselves):
Shallower curves for sure, but note the larger gap between red and green curves at all useable bass-mid crossover frequencies and a complete loss of coherence (around 75º at by particular crossover point of 2.2kHz) between mid and tweeter.
Incidentally, the group delay figures for this system are well below the threshold of audibility at all frequencies for which audibility data figures including all of the up-front analogue filtering are both around 0.5ms at 500Hz. The figure diminishes from there on up. Group delay is of course much higher at very low frequencies and a 16Hz- capable system such as this is somewhat susceptible despite being of sealed enclosure type.
As an exercise, I extrapolated from the data Anyway at 60Hz, the period T = 16.7ms. So that’s the threshold of audibility if extrapolating the available data is valid. Including all influences from the LT circuit (which actually improves things a bit), the LW4 crossovers, the 4th order Butterworth subsonic filtering, the drivers and the cabinet, the group delay of the bass system at 60Hz simulates at just 6.3ms.
Addendum: Here is a graph showing the above simulated group delay curves all under the Blauert & Laws points and my own 1-cycle extrapolation (click for full size):
Note: Actual in-room group delay was measured and is now at the DSP or NOT page.
Testing with the oscilloscope was very encouraging. All traces taken from the outputs at the crossover frequencies overlapped almost perfectly. 😀
Back to main speaker project page if interested.