Measuring distortion with software

The case with the “Chinese” Quad 909 was the first occasion we worked with a software based distortion analyzer.

I used to work with a Sound Technology 1710A. The unit needs calibration and some repair; I have sent mails to ST, but no response until now. Just a calibration is a staggering 995 dollar or so, ex P&P, if they accept the work. Then I realized that even if they were willing to repair the unit, the performance will not be good enough to measure modern or modified vintage equipment. So we moved on.

Distortion analyzers like the ST are based on wideband AC RMS meters. First you measure the output signal of the amplifier under test, then you remove the test tone and measure the residue. Eventually you can calculate the distortion. This is off course distortion including noise.

The software distortion analyzers are based on a spectrum analyzer program and analog-to-digital and digital-to-analog converters. This type of analyzers measures the individual frequency components, the rest is “only” calculation!

Spectrum analyzers existed in the pre digital days, but they where very expensive.

The theory behind this is the fact that every repeating signal can be decomposed in so called harmonics and the basic frequency component. To measure amplifiers you also need a signal generator and an oscilloscope in (preferable) one software package. I started with a freeware solution from Sillanum Soft and the standard sound card in my PC. The results were very promising. The next phase was a high quality, 24 bit, 192 KHz sampling rate, low distortion, soundcard with balanced outputs (ESI Juli@). The results were more than ok. Because the Sillanum Soft package only has the basic functionality, I bought a package from Virtins Technology and built a proper brake out box with volume controls, a variety of connections and a possibility to connect load resistors and other lab equipment. To get more speed (calculation power) from the PC, I also installed a 64 bit operating system (dual boot option).

From the website of Audio Precision, a high end distortion analyzer manufacturer, I downloaded two very useful documents. The Audio Measurement Handbook and a large document titled "How to write and read audio specifications". By reading and digesting the last document you will find that most specifications of commercial audio equipment are not complete and meaningless in most cases! Even our Quad heroes wrote those kind of specs!

A practical case: measuring a Dada Electronics revised Quad 303

See the specifications in the picture. First I checked the distortion. Quad measured at a strange frequency of 700Hz. Also the term “unrestricted bandwidth” is undefined. What does ”up to 45 watt” mean in real terms of voltage and watts? My own settings: 1 KHz test signal, bandwidth: 20Hz to 20 KHz and 45 watt in 8 ohm loads, both channels driven and loaded. As you can see the specification of 0.03% is easily met. If you remove or lower the load the result is spectacular. Very low distortion indeed.

The next measurement was the crosstalk. Quad did not specify the output parameters. As you can see, with the channels loaded with 8 ohm and the power at 45 watts, the channel separation is below modern standards. Only good enough for vinyl replay or FM broadcasts. If you remove or decrease the load, it is ok.
So there are people who say the 303 is the best Quad amplifier ever made. We don’t agree if you use the 303 with modern digital source material combined with high volume settings. This is one of the reasons we advise to monoblock a 303 or use it in a bi-amp application per channel. The load dependent cross talk can influence the stereo soundstage.

The next myth: high bias current

In the Quad service manual it is stated that the DC bias current of the power transistors must be between 10mA and 5mA. A popular “mod” is increasing the current to 35mA and in some cases even higher. Well, in real life the distortion is only 0.05% at a bias current of 0.5mA. At 5mA it is 0.01% and will not get lower at higher bias currents. Also the spectrum analyzer will not give a better (cleaner) picture at high levels of bias current. So Quad was (is) right! For practical reasons, stability of the circuit, 10mA is a good value for the bias current.

Joost Plugge

Virtins Technology
Sillanumsoft
ESI
Audio Precision

Improving a new Quad 909

One of our French customers, who is working in professional audio, bought a new Quad 909 to drive his Tannoy studio-monitors.

He has always used Quad 303 monoblocks with Dada High-end boards and was very pleased with the result so, as he needed more power for another project he bought a 909.

The Tannoys are very sensitive and transparent (as studio monitors should be). When he connected the 909 he heard a hiss and distortion in the mid-tones.He brought back the amplifier to the reseller and it was tested by Quad. They said everything was working within specifications so there was nothing they could do.

So he sent the amplifier to Dada Antwerp and asked us to find the reason of the problem and to see what we can do about it.

Joost and me took an afternoon to do some tests with Virtins Multi-instrument Audio analyser and the other lab-equipment.

First we did a listening-test. The sound was rather "harsh" and "metalic" compared to other (revised) Quads.

We measured a Total Harmonic Distortion of 0,8854% (!) and a Signal-to-Noise Ratio of 81,4dB at full power with 8 Ohm load. The total power before clipping was 120W RMS. Very bad figures for a high-end amplifier... And much worse than the "classical" 606 the 909 is based on.

We decided to take the amplifier apart and to concentrate on 3 possible sources of distortion:

• The power-supply


• The amplifier-boards


• The input-board for the Quadlink

The power-supply board is the "classical" board known from the 606-II and the 707 with a Toroïdal transformer and 4x 15000µF capacitors in a dual-mono design. Only, the original BHC Aerovox capacitors have been replaced by Chinese CapXon capacitors.

We took them out to test them with the Hameg LCR meter.

The BHC Aerovox was the ALP22A 63Volt, the CapXon has no type-indication, it is a plug-in two pin 80Volt 105° type.

We measured the capacitance at 100Hz, 1KHz, 10KHz and 25KHz with a 1V DC bias voltage. Both were within specifications (15mF or more) up to 1KHz. At higher frequencies the BHC scores better with 3x higher capacitance, for example 1100µF at 10KHz instead of 300µF for the CapXon.

The measured internal resistance is about the same and around 10mR at 1KHz.

This leads us to the conclusion that both capacitors are very comparable with a slight advantage for the BHC specifically in the higher frequencies. Of course this test doesn't say anything about the ageing of the capacitors after several hours of usage. The BHC has an excellent reputation for ageing. We decide to replace the 4 CapXon in the power-supply with BHC Aerovox.

Then we take a look at the amplifier boards. We replace the low-cost capacitors as usual, the electrolytes with Nichicon, the range between 1nF and 1µF with Wima MKS and the range below 1nF with Silvered Mica.

We also recalculated the bridge and used the exact values for the bridge components. Joost will comment on this later.

We connect the driver-boards to the power-supply. Now we measure a THD of 0,025% and a S/N Ratio of 87dB which is a lot better yet. A listening test gives the rich and transparent sound we are used to with Quad amplifiers.

Finally we take a look at the input-board that also contains the switch-on delay and the symmetric -> asymmetric conversion for the Quadlink.

The asymmetric signal from the RCA-inputs also passes through the output-stages of the preamplifier and again we find some low-cost capacitors on board. At a low input voltage the THD of the board is 0,003% but increasing the input-voltage increases the THD of the board. Around 10V we have 0,4% THD.

As the customer won't use the Quadlink we decide to remove the board from the signal-line and to connect the RCA-inputs directly to the amplifier-boards like in the 606.

The conclusion is that every detail has it's importance for the final sound-quality, specially the quality of the components and that it is best to avoid the Quadlink if you don't need it.

Stefaan

Practical: Component list

The 909 PCB is a thhrough-metalised PCB. You should use a good desoldering-pump or a desoldering-station to remove the old components.

For soldering on the PCB's only use lead-free solder. Tin/lead solder won't mix with the solder on the board.

You can click on the components below to go directly to the Webshop for ordering or for seeing more technical information or price and availability.

• The 4 Power-supply capacitors on the central board are replaced with BHC Aerovox 15000µF 63V ALP22A or ALC10C.


• (2x) R29 (in the bridge) is replaced with 10R 1/2W 1% Metalfilm


• (2x) R30 (in the bridge) is replaced with 47R 1/2W 1% Metalfilm


• (2x) C1, C4 and C6 (6 in total) are replaced with 330pF Silvered Mica


• (2x) C7 is replaced with 100µF 63V Nichicon


• (2x) C9 and C11 (4 in total) are replaced with 470µF 63V Nichicon

This way we have replaced all electrolyte capacitors with the best quality and we have adapted the bridge to the optimum value.

The specifications of the 909 transformer are: 2x 80,5 Volt 500VA

Measurement of a 606 MKII (my own) as a reference:

Vout is 32.2 V, no load, left and right channel:

THD 0.0009 and 0.0010

IM Din 0.0045 and 0.0041

Vout is 32.2 V, load is 8 Ohm:

THD 0.0129 and 0.0059

IM Din 0.0342 and 0.04

So the left channel is out of spec, THD should be lower than 0.01 some work ahead!

Joost Plugge