Hi,

I've remeasured the triode-connected Trigon EF86 valve at currents below 1 mA and I've reprocessed the old recordings for 2 mA and 4 mA. The latest results are:

Flat, 20 Hz to 20 kHz:

199 uA: 16.45 nV/sqrt(Hz)

297 uA: 13.87 nV/sqrt(Hz)

891.5 uA: 9.045 nV/sqrt(Hz)

2073 uA: 7.384 nV/sqrt(Hz)

3960 uA: 8.541 nV/sqrt(Hz)

A-weighted, 20 Hz to 20 kHz:

199 uA: 15.73 nV/sqrt(Hz)

297 uA: 13.11 nV/sqrt(Hz)

891.5 uA: 8.438 nV/sqrt(Hz)

2073 uA: 6.482 nV/sqrt(Hz)

3960 uA: 6.357 nV/sqrt(Hz)

100 Hz to 300 Hz:

199 uA: 19.9 nV/sqrt(Hz)

297 uA: 20.02 nV/sqrt(Hz)

891.5 uA: 20.8 nV/sqrt(Hz)

2073 uA: 26.27 nV/sqrt(Hz)

3960 uA: 38.03 nV/sqrt(Hz)

1 kHz to 1.2 kHz:

199 uA: 16.32 nV/sqrt(Hz)

297 uA: 14.56 nV/sqrt(Hz)

891.5 uA: 10.01 nV/sqrt(Hz)

2073 uA: 9.062 nV/sqrt(Hz)

3960 uA: 8.871 nV/sqrt(Hz)

10 kHz to 12 kHz:

199 uA: 16.21 nV/sqrt(Hz)

297 uA: 12.89 nV/sqrt(Hz)

891.5 uA: 8.205 nV/sqrt(Hz)

2073 uA: 5.798 nV/sqrt(Hz)

3960 uA: 5.597 nV/sqrt(Hz)

Needless to say, I haven't a clue how accurate these measurements are.

At least you clearly see the trends: white noise gets less with increasing bias current while excess noise increases with increasing current. The total noise around 200 Hz stays more or less constant up to about 1 mA and then increases. The optimal current for audio use is in the range of 2 mA to 4 mA, presumably on the lower side of this range for RIAA amplifiers as they have more gain for lower frequencies.

The trend for 1/f noise below 1 mA was very different on the crappy Amperex devices, there it increased enormously below a certain current. Presumably this was the island effect that Van der Ziel wrote about.

My measurement procedure is now:

-Switch off all equipment that you don't need, preferably by unplugging it, to minimise interference

-Record the noise with the valve switched off

-Record the noise at various bias current settings with grid grounded

-Record the noise at various bias current settings with grid connected to a known resistor that is used as a predictable noise source (44890 ohm for the old measurements, 20 kohm +/- 1 % for the newer measurements)

-Use a comb filter in GoldWave to suppress 50 Hz mains hum and its harmonics (wave

-wave(n+882) at 44.1 kHz sample rate)

-Use a steep 20 Hz highpass to remove any subsonics related to mains voltage variations (20 Hz, steepness=20 in GoldWave, which actually means a 40th order filter)

-Listen very carefully to each recording and select a part without any noises due to microphony and thermal expansion, the neighbour's GSM telephone, ticks or anything else that is not noise

-Maximise the level and write down how much GoldWave has increased the level

-Measure the RMS level with GoldWave's "Volume Match" function. Volume Match determines the RMS value but neglects anything below a certain threshold, hence the need to first maximise the level to get an accurate result.

-For the spot noise measurements, bandpass filter the signal, maximise the level and write down how much GoldWave has increased it, cut off a small part from the beginning because the bandpass filter needs some time to settle, use Volume Match again to measure the RMS value. Don't make the measurement bandwidth too narrow if you have only short sound fragments left.

-Use some straightforward mathematics and a spreadsheet to correct for the measurement amplifier noise and to determine the valve noise by comparing it to the known noise from the resistor (assuming that the valve has only equivalent input voltage noise, no input current noise).

Best regards,

Marcel