This article, originally written June 2009
has been updated October 2010

My Penultimate Audio Project:
A bi-amplified stereo system using Prof. Marshall Leach power amplifiers

Part 1: System Design

The primary configuration objective was for stereo bi-amplified main speakers with low level crossover plus two subwoofers. The subs have their own amplifiers, which are driven from the low pass output of another low level crossover. I chose the sub/main crossover frequency as 62Hz. The main amplifier pairs are therefore fed with frequencies above 62Hz. Each main amplifier system has a crossover at 350Hz. The LF speaker carries frequencies from 62Hz to 350Hz. The HF speaker carries frequencies from 350Hz and up. All crossover points are 6dB down and are Linkwitz-Riley 4th order configurations.

The audio system block diagram can be found here.

My old Yamaha receiver still acts as preamp. I am also using it for driving the rear speakers when running 5.1 modes.

Bi-amplification has several advantages. With two equal power units, you get the equivalent of 4 x the power output of a single amp when equating non-linearity performance. That assumes you feed each amp with equal spectral power. This is one reason why I chose 350Hz as crossover frequency. This frequency is regarded as about the point which has equal energy below and above for most musical types. Other good reasons for bi-amplification include the elimination of a high level crossover network. The bass driver is connected directly to the LF power amplifier. By using low level crossovers, well controlled crossover characteristics can be obtained and level matching for the sensitivity of the LF and HF drivers is a trivial exercise.

My system is really tri-amplified, since the bass is filtered off to dedicated sub-woofers. This is good for relieving the stereo bass drivers of the hard work (below 62Hz in this case). Subwoofers may be positioned in the room to try and mitigate the worst room resonance modes and using multiple subwoofer units can improve this situation even further. For the moment I use one sealed subwoofer cabinet of 48 litres internal volume housing two ACI SV12 drivers. Yes, the drivers operate below resonance in this box and there is a Linkwitz transform equaliser to compensate for the 12dB/octave slope.

Choice of speaker drivers

The fundamental goals for driver selection was for tight low distortion bass and also to avoid that scourge of audio reproduction - sibilance. Vocal sibilants have to be managed right from the microphone and it is hard to mitigate once problems have occurred. Unfortunately there are many recordings where not enough attention has been paid to this by sound engineers -they can’t all be tone deaf! In terms of speaker drivers, good low 2nd order non-linearity in the 1kHz to 6kHz area is important. In addition, avoiding any response peaks or resonances about 5-6kHz will help by not over-emphasising vocal sibilants (ess-ing) as well as not over-emphasizing distortion components from fundamentals in the 1-3kHz region.

My original plan was to have a single bass driver from 62Hz up to 350Hz plus a single wideband HF driver operating from 350Hz up. Because the ear is very sensitive to speech frequencies, both linear and non-linear distortions are more audible in this region. I wanted to avoid crossovers in the speech band from 300Hz to 3kHz. Ultimately I chose 350Hz as the LF/HF crossover frequency. However, I could not locate a single HF driver suitable for 350Hz to 20kHz. There are quite a few that claim to do this. Most I looked at had quite uneven response curves. So I went to plan B, which meant a midrange plus tweeter combination and then, of course a high level crossover from mid to tweeter. I chose a 5kHz crossover frequency. At this frequency, power levels are low and losses are no great problem. Furthermore a first-order network seemed possible thus avoiding major group-delay issues. More details of the speakers later in the article.

Part 2: Amplifiers

The power amplifiers are designed by Prof. Marshall Leach and are widely known as the ‘Leach amplifiers’. I built two of these into each of two chassis’. That’s four boards in total. One chassis drives the left speakers and the other chassis drives the right speakers. One P.A of a chassis drives the LF speaker and the other P.A drives the HF speaker. I used an active crossover which is a 4th-order Linkwitz-Riley configuration. The crossover was designed by Rod Elliott. I made the crossover frequency 350Hz. Input to the crossover board is via a balanced to unbalanced line receiver from Elektor. This is the line input to the amp chassis and itself is driven from a separate interface unit which splits off the low bass for feeding to subwoofers. Each chassis unit receives frequencies from 90Hz up. See the amplifier chassis block diagram here.

Here you see the 625VA toroidal transformer with separate secondary windings and independent rectifier/capacitor input filters for each amplifier. This ensures that any supply modulation due to current draw from one amplifier has minimal effect on the supply rails of the other amp. A small auxiliary transformer feeds the control board and the crossover section.

A large tightly coupled toroidal transformer has a significant inrush current. The primary dc resistance of this one is 2 ohms. The on-off system incorporates 2 power relays. The first is the main on-off relay and it feeds AC power to a 10 ohm resistor in series with the transformer. The second relay bypasses the 10 ohm resistor after 250 milliseconds. The relays are controlled by a control board which incorporates the delays and provides the single button on-off switching. There are also thermal sensors on the P.A heatsinks, which would switch off the amplifier in the event of abnormal temperatures.

Construction

Have a look at one of the built chassis. Rather industrial looking, I’ll agree. Nonetheless, they hide away in a cabinet largely unseen.

chassis small

The main photo was taken before I added the crossover circuitry on the right hand side. This addition can be seen here.The screened box houses the balanced line receiver, 350Hz crossover section and a low power balanced 15V regulated PSU.

Construction of the power amplifiers is well documented by Professor Leach but is not really for a novice. Transistors and zener diodes require matching and the power transistors must be carefully mounted on heatsinks along with the bias compensating diodes. There is a pc board negative available and I understand there is now a board supplier. I would consider it unwise to redesign the board or stray too far from the general construction methods given. Supply rails are 58V balanced. Supply lead dress is important for best performance and it is vital to follow the star-point ground wiring methodology.

Part 3: Amplifier Measurements

This page continues with measurement results. Testing was done using an Audio Precision System One.

Power Amplifier Frequency Response: Within 0.1dB between 20Hz and 25kHz.

Response Plot

This graph shows the response of both Leach power amp boards in one chassis.

Power Amplifier Total harmonic distortion & noise: Under 0.1% at all power levels between 1 watt and 130 watts over frequencies 20Hz to 10kHz.

THD&N Plot

The graph shows THD&N at four power levels into 8 ohms. At low power levels the result is dominated by noise and some 2nd harmonic. The analyser result includes harmonics up to 80kHz. The Leach amp is less than 3dB down at this frequency and so the non-linearity is fairly represented by measurement of harmonics. In fact a spectrum analyser revealed only 2nd harmonics. Higher order harmonics could not be seen.

Other measurements

Many measurements are required to fully characterise an amplifier. I have not made every measurement possible; but enough to convince me that the amps are behaving as they were designed. Some of the other tests I did make are continued below.

For those who regard single tone distortion tests as unrepresentative, a 2-tone test using equal level tones around 10kHz spaced 80Hz apart, gave a 2nd order difference tone (80Hz)IM level of 0.0046%  [-87dB]relative to one incident tone. Each tone alone resulted in 34 watts output.

Noise level between 400Hz and 22kHz was 68dB below 1 watt.

Hum components: For this test, I used a 20kHz input signal and ran the amplifier at 10 watts output power. Then I measured hum components relative to the 20kHz output.

  • 50Hz: -94dB
  • 100Hz: -102dB
  • 150Hz: -105dB

The hum test is important to ensure the PSRR is satisfactory and that lead dress and earthing techniques are optimised.For you 60Hz people, I expect the measured values to be similar, all else being equal.

In the next section, I note the difficulties I encountered with the overall system.

Part 4: Issues with the amplifiers

There were 3 issues encountered during construction and testing. None of these were attributed to the Leach amplifiers themselves and I consider them very successful. Rather the issues are ones of configuration and finger trouble.

Blowing up a Leach

When doing performance measurements using the System One, one amp suddenly stopped and a burnt resistor smell filled the lab. We never did find exactly why this happened, but four output transistors, two resistors and two capacitors later, the board was working again. Speculation goes that during testing the heatsinks became extraordinarily hot, the output transistors shorted. One supply fuse blew slightly before the other. Reverse voltage took out a supply cap on board, which then took out the small series resistor feeding the low level stages. To try and prevent such an event reoccurring, I added reverse diodes across the main supply bus after the fuses, reduced the fuse rating slightly and added thermal switches to the heatsinks to shut off the amp. I suspect such an event would never occur in normal operation; the heatsinks have only ever become slightly warm even with extended high power operation.

Burping

Another board was built and extensively tested. When the remainder of the chassis was completed and first powered up, I could induce startling intermittent burping sounds from the board when it was tapped. I removed it from the chassis and with an eyeglass, found I had not soldered one pin of a transistor on the board at all. How it made contact happily for hours of testing will remain a mystery.

Buzz

The Leach amplifiers alone are very quiet. However, after I had built the crossover system into the chassis and wired the respective outputs to the inputs of each amplifier, there was a significant buzz present on the outputs of each. The performance of the amplifiers, the crossover and input line receiver alone were exemplary and could not account for the effect. It was not hum; there was very little 50Hz and 100Hz. The buzz comprised AC mains harmonics starting at 150Hz and extending up to nearly 1kHz. I have managed to reduce this buzz to nearly inaudible by a lot of work optimising earthing the boards and connectors as well as screening the line level stages. At one point I was blaming radiation from the main filter capacitors!

At this stage, I had a working system until I connected the audio input from the preamp. The inputs are balanced interconnects using balanced audio cable about 10 metres long and XLR connectors . Then some of the buzz became audible again although not as bad as originally. I really started losing hair when I realised the buzz became worse during evenings. Finally, it dawned on me that the buzzing noises were worse when the TV was on! The long interconnects run past the TV set and the amplifiers are plugged into the same AC socket as the TV. The current solution involved some more earthing of the preamp and sub interface unit and not running the amplifiers from the TV outlet.

The buzz was not cancelled by the balanced interconnection; rather the problem is related to earthing of the various AV units in the lounge. I think the large toroidal transformers have a part to play in the problem because they are so closely coupled that they are effective in coupling AC mains harmonics despite having an electrostatic screen.

I need to do more investigative work in this, but for now, the system is working and I can only hear very faint buzz with an ear right up to the speaker.

Part 5: The Speakers

I built the LF speaker into one cabinet and built another cabinet for the mid/treble speaker. The HF cabinet sits on top of the LF cabinet. The cabinets are sized so that when I am seated, my ears are near to the same height as a point between the tweeter and midrange driver. Also I wanted to be able to be able to assess any time alignment issues by moving the HF cabinet relative to the LF cabinet.

Here is a photo of the completed speakers.

The bass driver is a Seas W22EX001. I chose this mainly from the work of Seigfried Linkwitz who has looked at both linear and non-linear distortions of various drivers. Originally I chose the volume of the closed box to create a box Q of 0.8 with two of these drivers. However, in the end I used only one driver since I was going to filter off bass below 62Hz anyway. This would have made the box Q rather low, perhaps 0.6, since I didn’t alter the box size. However, by using more internal bracing to reduce the volume, the final box Q became about 0.7. It is maximally flat although that isn’t an issue since it handles nothing below 62Hz. The driver receives frequencies between 62Hz and 350Hz (at the -6dB points)and is very flat over this range. The bass resonance of the driver when in the box is measured at 48Hz.

The midrange/treble cabinet

It is widely accepted that the midrange is the most critical frequency band and also the most difficult to engineer. The ear is most sensitive to frequencies in the midrange particularly over 500Hz to 3kHz. I wanted to avoid any crossover networks in this range and chose to use two Vifa P13WH-00-08 midrange drivers (in parallel for power handling.)

The two mid-range driver selection criteria I used were that there were to be no peaks in the band 4-6kHz to avoid accentuating vocal sibilants and that the dispersion off-axis was to be sensibly smooth. This Vifa driver meets these criteria and it was one of a very few I could find available that did so at that time.

Crossover to a Seas 27TFFC tweeter is at 5kHz. This tweeter is also a driver with smooth off-axis response and was available at a reasonable price. The first passive crossover I tried was first order but was later redesigned so that the low pass section is 2nd order and the high pass is 3rd order. New crossover diagram is linked on Part 6 below

Inductors are air core and capacitors polypropylene.

The cabinet baffle is an isosceles trapezoid in shape. The two midrange drivers are at the bottom and the tweeter at the top, mounted as close together in the vertical axis as was practical. This is not a MTM configuration. I sacrificed the benefits of MTM in order to make the cabinet as narrow as possible, aesthetics notwithstanding, so to try and manage diffraction. I reasoned that my listening position is fixed relative to the height of the drivers and the height of the cabinets is designed to suit. Diffraction (at midrange and higher frequencies)can also be made worse by knife-edges such as wood frames to support grille cloth. I cut some medium density foam to the same shape as the baffle, with cutouts for the speakers, stretched the cloth over the foam, then glued the whole lot to the speaker baffle. The foam may assist with reducing diffraction, but in fact there is very little that can be done about the sudden pressure drop that the initial pressure wave from the driver encounters at the cabinet edges. Because of the shape of my baffle, one midrange driver will have a slightly different diffractive effect to the other and this may help to soften the worst of the effects.

Impedance

The impedance of the bass speaker is shown here. Minimum magnitude is 6.6 ohms. Greatest phase angle is -42 degrees. Over the operating range of 90Hz to 350Hz, phase angle of impedance ranges from -28 degrees to +5 degrees.

Impedance plot of the mid-treble speaker is shown here. It is a benign load over the operating range of 350Hz to 30kHz. Z magnitude ranges from 3.4 ohms to 5.6 ohms and the angle ranges from -10 degrees to +15 degrees.

Acoustic Measurements

Some preliminary acoustic measurements have been made, however I intend to repeat these to assess the measurement sensitivities to the test environment and ensure I interpret the Praxis results correctly.

Next section: Initial auditions of the complete system

Part 6: Initial Audition

At this stage I have done only a few hours of not so critical listening. My first impressions are that midrange and vocals are smooth and are certainly without objectionable sibilance. In this respect my system is better than my old Energy Pro 4.5’s. The sound is spacious and doesn’t change character too much when I go to adjoining rooms. This is a good sign, indicating that room reflection spectra are not too different from the direct spectra. I have yet to assess the bass end properly although tracks from ‘The Wall’ album on CD do sound dynamic with pleasing impact. I’ll leave it here for now and hope to expand on this section and include some in-room measurements before long.

To be continued

Part 7: Update Sept 2010

FINALLY I have been able to do some in room measurements. It has taken me over a year to get around to this. Using the Praxis system and an ECM8000 microphone I have taken some in-room measurements.

The results were not as good as I had hoped for and reflect my subjective impressions that vocal intelligibility was lacking. Specifically, the frequency response on-axis had a deep hole of about 9dB in the 3kHz to 4kHz area. This speaker configuration with a single tweeter above two mid range drivers is prone to VRP variations making the response dependent on vertical position. I had aimed for a flat response at my seated head height which is on a line with the upper mid driver. The response changes at different heights but always included a dip at 3-4kHz.

Having discovered this, I had another look at my crossover and made empirical alterations. The new passive mid/high crossover network is here.

The original problem dip was being caused by interaction between tweeter and midrange networks and so I had to abandon the lofty first order crossover ideas. I have managed to reduce the dip now to under 5dB and shift the frequency of the dip to 5kHz, which has improved voice reproduction considerably. Still not perfect but a lot better. The response is now generally within 3dB from 300Hz to 20kHz excepting that one problem around 5kHz. (Noting that this speaker is electronically crossed over at 330Hz with LF duties taken up by an 8″ woofer.) I clearly have more work to do but as readers will know, these things take a lot of effort if one is trying to run a business and home life as well. See the old and new on-axis responses here.

I have not completed off-axis measurements, first wanting to improve the response further. My initial crossover designs were made using simple electrical models only and the results do show up the difficulties with those techniques. I do not have sophisticated speaker modelling software. Impulse response and step responses are not as classical as I would like but these are most likely due to the conventional use of drivers as I have done. A lot more work will be needed to correct this situation but let us see if it takes another year….