If paralleling a few transistors results in lower noise, is the limit to how many can be paralleled just the space required? Is there a point of diminishing returns?
Transistor count should double for next 3dB improvement. At some point the whole thing gets unstable due to cumulative inter-electrode capacitance.
If you put "N" transistors in parallel, you effectively put N intrinsic base resistances rbb in parallel. The rbb of the parallel assembly is the single transistor rbb, divided by N.
Since voltage noise is proportional to the square root of resistance, the noise of N transistors in parallel, equals the noise of a single transistor, divided by sqrt(N).
4 transistors in parallel --> half the noise
9 transistors in parallel --> (1/3) the noise
16 transistors in parallel --> (1/4) the noise
100 transistors in parallel --> (1/10) the noise
900 transistors in parallel --> (1/30) the noise
Since voltage noise is proportional to the square root of resistance, the noise of N transistors in parallel, equals the noise of a single transistor, divided by sqrt(N).
4 transistors in parallel --> half the noise
9 transistors in parallel --> (1/3) the noise
16 transistors in parallel --> (1/4) the noise
100 transistors in parallel --> (1/10) the noise
900 transistors in parallel --> (1/30) the noise
I read that the National Semiconductor LM395 paralleled 50 transistors for each transistor in the pair.
http://www.ti.com/lit/ds/symlink/lm394.pdf.
http://www.ti.com/lit/ds/symlink/lm394.pdf.
This is valid for noise voltage. The noise CURRENT gets *worse*
by the same numbers.
The noise current flows through the source resistance and
produces there a noise voltage also, just according to Ohm's law.
If the source resistance is large, that can be the governing part.
But if you are after small voltage noise, the source resistance
has to be small anyway. A 60 Ohm resistor produces 1nV/sqrt(Hz)
thermal noise voltage already at room temperature.
So, for every source resistance there is an optimum number of
a given transistor. This is usually not infinity.
not enough transistors --> too much voltage noise
too many transistors --> too much current noise.
Noise voltages of several sources add geometrically. (root of sum of squares)
regards, Gerhard
Last edited:
LM194/394 ?
Maybe you write about National LM194/LM394 supermatch pair (I used it in the past on an Jean Hiraga RIAA preamp design). It was a good substitute for input transistors (usually used in differential configuration) of opamp like LM318; NE 5534; TL 072/82 etc.
Maybe you write about National LM194/LM394 supermatch pair (I used it in the past on an Jean Hiraga RIAA preamp design). It was a good substitute for input transistors (usually used in differential configuration) of opamp like LM318; NE 5534; TL 072/82 etc.
Hi,
The LM394 was and can be used in MC amplifiers to standard phono.
It has no advantage over the 5534 for a standard phono RIAA stage.
rgds, sreten.
The LM394 was and can be used in MC amplifiers to standard phono.
It has no advantage over the 5534 for a standard phono RIAA stage.
rgds, sreten.
Cicero32 is correct. I meant the LM394.
Just imagine what paralleling all those transistors does the base-collector capacity and the huge Miller effect capacity you would have with any reasonable amplification. The 2SK147 FET had the same problem. It had great transconductance obtained by using a large interdigitated gate structure and this resulted in 17pF of gate-drain capacitance. This caused big Miller effect.
Just imagine what paralleling all those transistors does the base-collector capacity and the huge Miller effect capacity you would have with any reasonable amplification. The 2SK147 FET had the same problem. It had great transconductance obtained by using a large interdigitated gate structure and this resulted in 17pF of gate-drain capacitance. This caused big Miller effect.
Correctly biased is more silent than NE5534 input transistors; in the past it was used as substitute of input transistors of many opamps; 5534 included; specially here in Europe!
The Hiraga design had (in 1982) nearly the same spectral density of HP Spectrum Analizer except for 1/F noise on lowest frequencies.
The Hiraga design had (in 1982) nearly the same spectral density of HP Spectrum Analizer except for 1/F noise on lowest frequencies.
Hi,
With typical MM cartridges adding the LM394 to a 5534 is pointless.
By all means add it for a MC input with lower source impedances.
rgds, sreten.
With typical MM cartridges adding the LM394 to a 5534 is pointless.
By all means add it for a MC input with lower source impedances.
rgds, sreten.
LM194/394 noise.
AD797 with it's 0,9nV/SQR Hz was (in 1982) till in the mind of his designer!
LM394 not was added to OPAmps it was used in SOSTITUTION of OPAmp's
input transistors (usually inputs was relied to V- blocking input transistor and LM394 relied on compensation pins, becoming REAL input transistors!). With a little selection and polarizing with a slighty different current value, we had 1,6 nV/SQR Hz too!
From datashet:1,8nV/SQR Hz for LM194/394 3,5nV/SQR Hz for NE5534
AD797 with it's 0,9nV/SQR Hz was (in 1982) till in the mind of his designer!
LM394 not was added to OPAmps it was used in SOSTITUTION of OPAmp's
input transistors (usually inputs was relied to V- blocking input transistor and LM394 relied on compensation pins, becoming REAL input transistors!). With a little selection and polarizing with a slighty different current value, we had 1,6 nV/SQR Hz too!
Hi,
Nonetheless the better numbers are pointless with a typical MM for a 5534.
A typical MM has 1K source resistance and significant inductance, such
that the source impedance rises to to about 47K within the audio band.
The LM394 can be added for MC inputs, and was and still can be.
rgds, sreten.
I recall reading years ago that the the 5534 paralleled the 5532
input transistors to reach its lower noise figure. I don't know
if its true, but it made perfect sense at the time.
Nonetheless the better numbers are pointless with a typical MM for a 5534.
A typical MM has 1K source resistance and significant inductance, such
that the source impedance rises to to about 47K within the audio band.
The LM394 can be added for MC inputs, and was and still can be.
rgds, sreten.
I recall reading years ago that the the 5534 paralleled the 5532
input transistors to reach its lower noise figure. I don't know
if its true, but it made perfect sense at the time.
Last edited:
-
In the Hiraga pre the the difference behind shortcircuited and open circuit was + 15dB (S/N shorted input = 89,8 dB S/N open input =75 dB) mainly due to 47 kOhm load resistor and the input current flowing throught it! The purpose of this design based on LM194 and LM118 (S.R. 70 V/uS op amp), was to have a pre with a silent input stage without degrading slew rate.We had a 85 dB S/N with input relied to Shure V15III pickup (measured with an HP spectrum analizer), there was also a cascode fets input but was useless (considering the noise of input resistor). The pre was very accurate and musical and very very selective. It was able to appreciate even the smallest details, even the tricks used in the recording studio, making perfectly audible pumping effects due to dynamic expanders, DBX and so on.
Feedback is an issue, you need a high current output stage to properly drive a low value feedback resistor for low noise. The Borbely MC stage used a two gain stages. A flat 26 or 32 db first gain stage then the HF RIAA roll off and the 35 db second stage with LF equalization. Each gain stage runs at 100 ma to drive low impedance feedback resistors. My particular Borbely runs 4 x 2sJ 147 and 4 x 2SJ72 as the input stage cascoded. The equivalent input noise is that of 5 ohm resistor......
I recall the John Hardy 990 discrete op-amp uses (or used) the LM394. I didn't know it (the 394) was obsoleted, but from the link below, the 990 lives on with different parts. Here's the data sheet/history/blurb on the 990:
http://www.johnhardyco.com/pdf/990.pdf
http://www.johnhardyco.com/pdf/990.pdf
global NFB reduces the distortion and noise and output offset of the open loop amplifier.
There may be others in the list that I don't know about.
There may be others in the list that I don't know about.
- Status
- Not open for further replies.