Saturday, 26 April 2014

Hissssss and Pinggggg

When designing a tube mixer you always aim for as little noise as possible from the microphone pre-amplifier. The  inherent noise in tubes makes it very difficult to design a low noise mic pre-amplifier without an input transformer. Even with a 1:10 ratio input transformer, it is difficult to achieve the very low noise levels that can be achieved by the best semiconductor designs. One reason for  this is that the noise generated by tubes can vary a lot from tube to tube due to small  manufacturing differences. For this reason it is usually necessary to select the input tube of a mic pre for lowest noise.

In addition to their inherent noise, tubes have one other drawback when used for mic pres and that is microphony. When you tap them, an audible sound can sometimes be heard. Again, the degree to which this happens varies a lot from tube to tube of the same type so again it is necessary to select tubes for low microphony.

Now I have cured the hum issues with the demo mixer I am in a position to measure the noise of the mic pre and also to test various tube types for noise and microphony. I used the Helios 69 input channel with the EQ switched out. First I fed in a signal at -60dBu into the microphone input , set the gain to maximum (60dB), the channel fader fully up and then adjusted the second stage gain for 0dBu at the channel direct output. This means there was then a total of 60dB of gain from the mic input to the direct out. To measure noise I then plugged in a shorting XLR with pins 2 and 3 shorted,into the mic input. I then used my Lindos audio test set plugged into the direct out to measure the noise levels at the output.

A brief aside is necessary here to understand the results that will be presented. Noise in microphone preamps is usually evaluated relative to the thermal noise generated by the resistance of a typical microphone that might be plugged into it. A typical mic has a source resistance of around 150 ohms and this seems to be the standard by which mic pres are measured today. The rms thermal noise due to a 150 ohm resistor at 20 degrees C in a 20KHz bandwidth is close to -131 dBu. If we amplify this by 60 dB then the noise we measure at the output should be 60dB higher or -71dBu. The noise measured will in fact be higher than this due to the additional noise generated by the mic pre itself. If you measure the rms output noise and subtract the gain then you get the noise level at the input of the mic pre. This noise level is called the equivalent input noise or EIN for short. In a perfect mic pre with a 150 ohm source, the rms  EIN should be -131dBu. In a real mic pre you would expect it to be a few dBs higher.

Notice I said the 'rms noise' from the resistor. Most measuring systems cannot directly measure rms noise. The simplest are average reading and are calibrated to read the correct rms value when presented with a sine wave, but noise is not a sine wave so these can produce inaccurate readings for noise. In addition, weighting curves are often added to the output signal supposedly to more accurately reflect the response of the ear. The A weighting curve is one of the most popular and is often favoured by marketing departments because it tends to improve the EIN figure by a few dB. There is an international standard for measuring noise (ITU-R 468) which includes a curve that adjusts for the response of the ear and also includes a quasi-peak detector to account for the ear's sensitivity to short, spiky bursts of noise that simply would not show up on an rms reading device.

The bottom line is that the value of EIN you get depends a lot on the type of measuring device you use and the weighting you apply to it. This article by the Institiute of Sound and Communications Engineers lists the various methods in common use and the differences in the the results obtained. Many manufacturers quote EIN figures without giving any further information about how the measurement was made and what weighting was applied. It is not surprising that many of these quote an EIN close to -131dBu. The more honest ones achieve figures in the -128dBu region. For comparison, back in the 70s, Neve used to quote the EIN of its mic pres as -126dBu.

The tests I conducted used my Lindos test set which meausres noise to ITU-R 468 including the quasi-peak reading meter. This gives a more realistic figure for EIN but it can be anything from 4 to 7dB worse than the figure obtained using the simpler and conveniently weighted measuring methods. If you look at the published specifications for classic tube mic pres you will generally find that they quote an EIN figure of 'better than -120dBu' but usually without specifying how the measurement was made. We would like to do better than this. In the EZ Tube Mixer mic pre, the noise is determined principally by the 12AX7 tube used at the input. I tried three different types of 12AX7 in my tests:

  • 12AX7LPS made by Sovtek in Russia. I tested 8 of these.
  • 12AX7EH by Electro Harmonix and also made in Russia. I tested 9 of these.
  • 12AX7WA another variant made by Sovtek. I tested 11 of these.
All three types of tube are in current production. Each tube was tested in the same channel amp at 60dB of gain. To beat the -120dBu EIN of the classic tube mic pres we need to achieve an output noise level of better than -60dBu. After measuring the noise level I tapped each tube with my finger nail to see how microphonic it was. The Lindos test set has a built in loudspeaker preceded by a 50dB amplifier when it is making  noise tests. So there is a total of 110dB of gain before the speaker. This means you can clearly hear the hiss when making noise measurements and you can easily hear your finger tapping on the tube. With this test I divided tubes into three categories.
  • Microphonic - these tubes were judged to be so badly microphonic as to be unusable. With tubes in this category you got a very loud pinging noise when they were tapped and you could clearly hear your finger gently rubbing the glass envelope of the tube.
  • Good - these tubes produced an audible ping but it was less than 20dB above the noise level
  • Very good - these tubes produced a barely audible ping which was less than 10dB above the noise level

All of these are of course subjective measurements which is why I did  not feel able to split them into more than three categories. Note that none of the tubes was completely lacking in microphony. Every tube tested was microphonic to some degree. The results were as follows:

12AX7LPS - 6 of the tubes measured -63dBu for noise; one measured -62dBu and one was -58dBu; 2 tubes were microphonic, 5 were good and 1 was very good.
12AX7EH - 2 of these measured -63dBu for noise and  7 measured -62dBu; 8 had good microphonics and 1 was very good
12AX7WA - 9 of these measured -63dBu for noise and 2 measured -62dBu; 1 was microphonic, 2 were good and 7 were very good.

In conclusion, the 12AX7WA was the best all round with many measuring -63dBu noise level and plenty of very good from the microphonic point of view.  Both the 12AX7LPS and the 12AX7EH turned in  good noise performances at around -62dBu to -63dBu but their microphonic performance was not as good as the 12AX7WA.

I think the 12AZ7WA is the best choice from the noise and microphonics point of view. This is interesting because the internal construction is much shorter than the EH and LPS versions so perhaps this makes them more rigid and accounts for their better microphonic performance. What is also interesting is that I had previously discounted them for use as mu followers because they had rather higher levels of intrinsic distortion compared to other versions of the 12AX7, the LPS in particular producing very low levels of distortion.

The EZ Tube Mixer mic pre topology uses an SRPP output stage which I expect to be the main contributor to the overall distortion level. The 20dB or so of negative feedback in the mic pre is designed to reduce the SRPP output stage distortion by tenfold. This means I do not expect the distortion of the input stage 12AX7 to make much of a contribution to the overall mic pre distortion. Just to make sure, I measured the 1KHz distortion for the WA, LPS and EH types fitted in the first stage. The distortion measurements were made with the gain set to 30dB and with an input level of -20dB. This means the input to the 12AX7 is 0dBu and the output from the direct out transformer is at +10dBu. This means the SRPP output stage is running at +16dBu since the transform steps down by 6dB.

The results were remarkably consistent. The LPS type produced 0.094% distortion, the EH type produced 0.096% and the WA produced 0.094% distortion. This simply confirms that the vast majority of the distortion occurs in the output stage as expected.

It seems that the best noise, microphony and distortion is obtained with selected tubes of the WA type. The typical -62dBu measured output noise represents and EIN of -122dBu. This meets our target of better than the classic tube mic pres and if we add only the minimum 4dB to convert this measurment into its rms equivalent we are achieving an EIN in the region of -126dBu which is as good as a classic Neve.

Saturday, 19 April 2014


There comes a time in the build of almost any audio DIY project when you encounter a hum problem and the EZTubeMixer is no exception. After fixing channel 3 it was time  to make some measurements of each channel. I fitted channel 1 and channel 2 with the two complete channel amplifiers and tested them using my Lindos test set. Frequency response and noise level were both good. I then plugged in the other two channel amps. As soon as I powered up and the tubes had warmed up there was a loud buzz at about -30dB from the channel I had last tested. Looking on the scope the buzz was at exactly 100Hz but half the waveform was dead flat and the other half was a high frequency. I could not trigger well enough on that portion of the waveform to see exactly what frequency. I disconnected the HT supply and I could plug in all 6 modules and it was fine which at least demonstrated it was probably not the regulated 12V supply hooting. I took out all the first stage tubes (12AX7), plugged in all six modules and there was no oscillation so it was either getting into the first stage or the first stage was creating it. With four modules fitted and only one fitted with its 12AX7 the oscillation occured. I added a 10K across the HT to pull another 30+mA from the HT supply and repeated the above test. No oscillation. Adding a 12AX7 to a second module the oscillation returned.

Grounding is always a potential cause of hum problems. The power supply for this mixer is situated at the front and the dc supplies travel by cables to the motherboards at the rear. I needed to make a connection from the HT 0V at the power supply to the mains safety earth tag which is next to the mains inlet connector at the rear. My normal rule is to take a wire direct from the HT- to the  safety earth tag but, as the end of the HT run through the motherboards is right close to this tag I decided to connect from the motherboard HT 0V to the safety earth tag. I did not think it would make any difference. However, when checking the heater elevation voltage at the heater supply I connected a DVM from the heater -ve to a different earth connection that is used for the screen between windings in the HT transformer. With just two 12AX7s fitted, connecting the DVM to make this measurement stopped the interference dead in its tracks. Aha, I thought, grounding problem. So I disconnected the lead from the motherboard HT 0V to safety earth and made a connection direct to the the PSU PCB HT 0V. Switched on and it was dead quiet. So I added a couple more 12AX7s and tried again with 4 modules. I was very disappointed to find the buzz had returned.So, just to make sure it was not a problem with my Lindos test set, I unplugged it and tried again. With all four channels fully populated with tubes there was no buzz. Plugging in the signal source from the Lindos increased it a little but plugging in the output brought it back fully. I disconnected the mixer output from the the Lindos and plugged in a passive VU meter - and the buzz returned. With both the input and the output connected there was full buzz but with just one there was only a small buzz. This happened to all four channel amp line inputs and direct outputs. Puzzling but at least I was getting some sort of a feel for the cause.

I then added the last two modules (the bus amps) but was again disappointed to find that with absolutely nothing connected to the mixer, the buzz was there. So it was nothing to do with anything connected to the mixer, it was the mixer itself. So, thinking it was probably the channel amps themselves that were oscillating, I ordered some 33pF capacitors to place across the 47K feedback resistor to see if that got rid of the high frequency oscillation.

Just to be absolutely sure it was not the linear heater supply causing the problem, I temporarily replaced it with a switched mode power supply intended to supply LED lamps. It is rated at up to 15 amps so it should be able to cope with the heater inrush current. Sure enough, it coped perfectly well with the heaters from all six modules but unfortunately the buzz was still there. At least I now know that this little SMPSU, which only cost £15, is perfectly capable of supplying the heater power for a small mixer.

 In the meantime I contacted my friend Holger and told him about the buzz problem. He very quickly got back to me to say he had had a similar problem which he had cured by adding 10nF decoupling capacitors across the HT supply on each two channel backplane PCB. I did not have any 10nF 400V capacitors to hand but I did have some 220nF ones. I quickly attached one across the HT supply on each of the motherboards, replaced all six modules and switched on. To my great relief, all sign of the buzz and high frequency oscillation had vanished. I plugged in my Lindos test set and still there was no sign of buzz.

I then re-connected the linear heater supply and checked that was OK. I also added a safety ground link from the panel on which the power supply is mounted to the mains safety earth. Lastly, I refitted the power supply to the mixer and repeated the tests. I am pleased to say that even with the PSU inside the mixer there is no sign of hum or buzz. It is so free of hum that I was able to measure the EIN of the mic pres and I found one of the 12AX7s was quite microphonic - looks like these will need to be selected by hand.

The only unanswered question is what caused the buzz in the first place? Since decoupling the HT supply at the motherboard cured the problem this suggests  it is an HT supply impedance issue. Possibly the inductance of the HT cabling and the power supply smoothing caps is to blame. The silly thing is, very early on in my career (over 40 years ago) I learned the importance of decoupling power supplies where they enter a PCB. I even had it on my standard list of things to look for in design reviews. What did I not do on the EZTube mixer mic pre board?? Fortunately the motherboard decoupling does the trick but for future versions of these boards I think there is going to be on board decoupling.

Output Transformer Fault

Now I have all the output transformers wired up it was time to wire them the the XLRs, plug the Molex KK connectors into the backplane PCBs and test each channel's direct out. A couple of hours soldering and it was time for testing. Channel 1 was fine, as were channels 2 and 4, but channel 3 had a very low output. A quick check of the secondary dc resistance revealed it was very different from the other channels. So I removed the transformer panels to check if I had made a wiring error, but channel 3 was wired exactly the same as the others. Further dc resistance checks on the primary and secondary windings revealed that the VTB2291 transformer was wired backwards. I had wired to the correct numbered tags so it looked like the tag panel had been put on back to front by the manufacturer. Fortunately I had a spare transformer so I removed the faulty one and replaced it with the spare one. Channel 3 now functioned as expected.

I removed the links from the faulty one and again checked the winding dc resistances which confirmed the tag panel had indeed been fitted the wrong way round. I contacted Colin at Audio Maintenance who quickly supplied a replacement transformer.