Microfonie en door emissie veroorzaakte brom (Techniek Radio/TV)

door MarcelvdG , 20.08.2011, 09:32 (4845 dagen geleden) @ Ouwe Schipper

Hallo allemaal,

Ik heb enige tijd geleden wat metingen gedaan van de ruis van een EF86 versus de instelstroom, nadat ik in het boek Noise van Aldert van der Ziel gelezen had dat de 1/f-ruis van buizen zowel bij hoge als bij lage stromen toe kan nemen en je daarom proefondervindelijk een optimum moet vaststellen.

Bij buizen van het merk Amperex die ik ooit op Koninginnedag gekocht had, nam de 1/f-ruis enorm toe bij stromen onder de 1 mA. Bij de buizen van Trigon (gekocht bij Radio Twenthe) waren er alleen geleidelijke veranderingen.

Met RIAA- en A-weging waren de resultaten van een van de Trigon-buizen:
199 uA: 16,46 nV/wortel(Hz)
297 uA: 14,24 nV/wortel(Hz)
891,5 uA: 10,19 nV/wortel(Hz)
2,073 mA: 10,29 nV/wortel(Hz)
3,96 mA: 11,33 nV/wortel(Hz)

Uiteindelijk heb ik voor mijn versterker maar een instelstroom van 1,75 mA gekozen: niet al te ver van het optimum voor de Trigon en ruim boven de 1 mA waaronder de Amperex-buizen zich misdroegen.

Hieronder staan nog wat resultaten zonder RIAA-weging en een beschrijving van de meetmethode, in het Engels omdat ik er eerder over bericht heb op een Engelstalig forum en te lui ben om het te vertalen.

Groeten,
Marcel

Noise measurements on four triode-connected EF86's:

Two EF86's of the brand Amperex, bought second-hand on Koninginnedag (Dutch national holiday), complete box with all kinds of valves for one euro with no guarantee of any kind

Two EF86's of the brand Trigon, bought at a shop (Radio Service Twenthe in The Hague)

For all four I listened to the noise while checking the A-weighted noise by reading off quasi peak-peak values with an oscilloscope with and without 47 kohm//1 Mohm connected between grid and ground.

The first Amperex EF86 showed very erratic behaviour: very strong low-frequency noise that sounded like a broken contact below a certain bias current level (of the order of 1 mA), reasonably normal sounding noise at higher current, grid currents up to 1 uA flowing out of the valve.

The second Amperex EF86 had no measurable grid current, but it also had strong low-frequency noise that sounded like a broken contact below a certain bias current level, reasonably normal sounding noise at higher current (1 mA or higher).

The Trigon devices were more well-behaved: the noise sounded normal at all current levels I tried.

For all four EF86's the A-weighted equivalent input noise voltage was at its lowest at the highest current I tried: 4 mA anode current.

For the second Trigon EF86 I did more accurate noise measurements. Using a home-made 100 times amplifier with A-weighting filter and a laptop, I recorded the noise of the valve, of valve plus 47 kohm//1 Mohm between grid and ground, and only of the 100 times amplifier and computer. Using GoldWave, I applied a comb filter and a steep 20 Hz high-pass to get rid of any remaining hum and subsonic supply voltage variations. I then measured the RMS values (using Volume Match in GoldWave) and calculated the equivalent input noise from the results. I later repeated the measurements below 1 mA because there were too many extraneous sounds in the recordings. For these repeated measurements I used a 20 kohm +/- 1 % resistor. I then also reprocessed the old recordings above 1 mA.

The common-cathode circuit was not in a shielded box and it was earthed through the water tap.

The results are (expressed as an equivalent white input noise voltage that would give the same integrated noise):

A-weighted noise for three valves measured with the inaccurate quasi peak-peak method:
Second Amperex EF86, the one without grid current:
228 uA: infinite (that is, valve noise is so high that the effect of the 47 kohm//1 Mohm is not visible)
1 mA: 8.26 nV/sqrt(Hz)
4 mA: 6.19 nV/sqrt(Hz)

First Trigon EF86:
220 uA: 12.08 nV/sqrt(Hz)
362 uA: 11.58 nV/sqrt(Hz)
2.16 mA: 6.44 nV/sqrt(Hz)
4.05 mA: 6.26 nV/sqrt(Hz)

Second Trigon EF86:
201 uA: 16.21 nV/sqrt(Hz)
347 uA: 14.76 nV/sqrt(Hz)
2.11 mA: 8.76 nV/sqrt(Hz)
3.89 mA: 7.64 nV/sqrt(Hz)

For the Amperex devices the low-frequency noise increased very rapidly below 1 mA. I didn't measure that, but I could hear it very clearly.

Second Trigon EF86, with the improved GoldWave signal processing of December 2010, recordings made on 5 December 2010 with 20 kohm +/- 1 % as a known noise source for currents up to 1 mA, older recordings of October 2010 with 44890 ohm as a known noise source used for the higher currents (the old recordings had too many sounds that were not noise for the low currents, but not for 2 mA and 4 mA):

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(n)-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 sounds due to microphony and thermal expansion, the neighbour's GSM telephone, ticks or anything else that is not noise, also check the GoldWave running FFT spectrum for any irregularities

-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).