Let me 'clarify' my personal opinion on making discrete audio gain modules.
As I have, for more than 40 years, I support making discrete circuity, WHEN a real advantage is shown or implied over a quality IC. However, this is not always the case.
For example, when Parasound asked me to build a phono stage ultimately called the JC-3, I was told NOT to use discrete jfets, because we were having such severe supply problems sourcing enough Toshiba jfets for the rest of our products. At that point, I reluctantly decided to use quality IC's and hope for the best.
Of course, I could have just gone back to bipolar transistors everywhere, like I started with in 1967, and still have a reasonable topology, but what is the point? Space and other considerations also limited my choices, initially.
Now, what did I find? First, IC's CAN be very quiet, almost as good as massed pairs of jfets. IC's are pre-trimmed at the factory for extremely low voltage offset, so SERVOS are not always necessary. It is almost IMPOSSIBLE to get similar performance without monolithic pairs of active parts, and they are getting hard to get.
Back 40 years ago, when I had to replace an IC that I found, with discrete, the best slew rate, low noise, and +/- 24V operation IC that I could find at the time, the HA911, only had a slew rate of 2.5V/us. Today, 20V/us is typical, thank goodness for progress. At the same time, the input noise dropped about 10 times from the early days.
When I started to make discrete modules for the GD, and ultimately, Mark Levinson, the jfets that are sometimes discussed here, like the j113 and the j175, were newly introduced. Siliconix also made another part, the J110 that truly could do about 1nV/rt Hz at 1KHz or so. The bipolar transistor with the lowest voltage noise at the time was the 2N4401, 2N4403 pair. These are modest, medium beta devices, and were not good over a wide range of impedances. At that time, a typical low noise device might have 2-4nV/rt Hz, mostly from extra base resistance.
Only in 1978, did the Japanese introduce improved versions to replace the 2N4401, 2N4403 in many applications.
Unfortunately, even though we went through a 'golden age' of individual and dual bipolars and jfets, for maybe 30 years, that time is over.
Yet, many IC's have improved enough to be viable. I realize that many here think that IC's sounded as good as the best discrete, and the difference is in our imagination, but I have never found it to be so. However, IF you want to build a discrete circuit with parts out of your junk box, you are going to be very disappointed. Better to use IC's, with DUE CARE using the best passive parts surrounding them.
As I have, for more than 40 years, I support making discrete circuity, WHEN a real advantage is shown or implied over a quality IC. However, this is not always the case.
For example, when Parasound asked me to build a phono stage ultimately called the JC-3, I was told NOT to use discrete jfets, because we were having such severe supply problems sourcing enough Toshiba jfets for the rest of our products. At that point, I reluctantly decided to use quality IC's and hope for the best.
Of course, I could have just gone back to bipolar transistors everywhere, like I started with in 1967, and still have a reasonable topology, but what is the point? Space and other considerations also limited my choices, initially.
Now, what did I find? First, IC's CAN be very quiet, almost as good as massed pairs of jfets. IC's are pre-trimmed at the factory for extremely low voltage offset, so SERVOS are not always necessary. It is almost IMPOSSIBLE to get similar performance without monolithic pairs of active parts, and they are getting hard to get.
Back 40 years ago, when I had to replace an IC that I found, with discrete, the best slew rate, low noise, and +/- 24V operation IC that I could find at the time, the HA911, only had a slew rate of 2.5V/us. Today, 20V/us is typical, thank goodness for progress. At the same time, the input noise dropped about 10 times from the early days.
When I started to make discrete modules for the GD, and ultimately, Mark Levinson, the jfets that are sometimes discussed here, like the j113 and the j175, were newly introduced. Siliconix also made another part, the J110 that truly could do about 1nV/rt Hz at 1KHz or so. The bipolar transistor with the lowest voltage noise at the time was the 2N4401, 2N4403 pair. These are modest, medium beta devices, and were not good over a wide range of impedances. At that time, a typical low noise device might have 2-4nV/rt Hz, mostly from extra base resistance.
Only in 1978, did the Japanese introduce improved versions to replace the 2N4401, 2N4403 in many applications.
Unfortunately, even though we went through a 'golden age' of individual and dual bipolars and jfets, for maybe 30 years, that time is over.
Yet, many IC's have improved enough to be viable. I realize that many here think that IC's sounded as good as the best discrete, and the difference is in our imagination, but I have never found it to be so. However, IF you want to build a discrete circuit with parts out of your junk box, you are going to be very disappointed. Better to use IC's, with DUE CARE using the best passive parts surrounding them.
perhaps. So far I havent seen anyone come up with a circuit topology that amplifies and is self correcting in dc offset. Why is that? Why isnt it integrated?
I suggested the servo but that was decades ago. I go off and do other things and come back to my beloved hobby and not much has changed in circuitry. Well IC have changed some but even those still have dc servo as an outboard appendange. What's that all about?! -RNM
Last edited:
True, but you get back higher gain in the first stage and ultimately lower, or at least not higher, input referred noise.
jan
AD744 with a discrete diamond buffer driven from pin 5 is worth a try.
Thanks Jan! Yes, I fished through email and found that it was On Semi that I'd seen first, although the link to the Electronic Products issue has long expired. And I remember that bit about the Sanyo on-chip 220mohm resistors that tended to blow up
EDIT: I also remembered the suggested circuits, which I immediately wanted to do differently
A cool thing (npi) would be to infrared-image the die and use that information. Certainly could be fast response
Brad
If I remember well it must work at very high closed loop gain to be stable; moreover its output impedance is fairly high, and both its gain and freq response are strongly affected by loading - nice topology, but imho is pretty useless as a line stage unless some isolation of the gain node is provided by an output buffer, or the output cascodes are operated at a much higher quiescent current.
L.
Last edited:
Sorry everyone. Post #391 was a glitch.
The IR detectors were dropped in here specifically due to the IR image of a die they can provide. See here:
You can’t realistically have 2/10 d. Celcius resolution temp. mapping of a surface divided into 25x25um areas and following the dynamic variations of the temp at ~1kHz of all these “pixels” with other methods.
All this at a very reasonable price and minimum complexity for a manufacturer or researcher.
But still very expensive for the hobbyist. (That’s why the talking turned to the IR spot thermometer).
This rule of thumb is the basic equation of thermography :
"Emissivity + Reflectivity = 1"
You want to maximize emissivity for to have correct IR temp. measurement. You have to minimize reflectivity to achieve this (shiny aluminum is very reflective).
A trick to achieve a spot with max. emissivity (~1 ) on any material, is to drill a blind hole on it with a depth > 2-3d.
IR temp. measurement into this spot is really very close to the true local temperature.
Brad, it sounds to me (especially the counter-effect cooling had on the intensity of the emission) as if it was actually IR emission not adequately filtered in front of the CCD.
Did they verify the wavelength?
Wavebourn
Raising the temp of the whole opamp, will worsen the dynamic temp. unbalance.
A specific subcomponent when variably heated due to it’s I*R will find it more difficult to get rid of it’s thermal energy through the hotter surrounding mass.
The delta temp. with another subcomponent (less I*R loaded) will thus increase.
George
The IR detectors were dropped in here specifically due to the IR image of a die they can provide. See here:
A cool thing (npi) would be to infrared-image the die and use that information. Certainly could be fast response
Brad
You can’t realistically have 2/10 d. Celcius resolution temp. mapping of a surface divided into 25x25um areas and following the dynamic variations of the temp at ~1kHz of all these “pixels” with other methods.
All this at a very reasonable price and minimum complexity for a manufacturer or researcher.
But still very expensive for the hobbyist. (That’s why the talking turned to the IR spot thermometer).
This rule of thumb is the basic equation of thermography
:"Emissivity + Reflectivity = 1"
You want to maximize emissivity for to have correct IR temp. measurement. You have to minimize reflectivity to achieve this (shiny aluminum is very reflective).
A trick to achieve a spot with max. emissivity (~1 ) on any material, is to drill a blind hole on it with a depth > 2-3d.
IR temp. measurement into this spot is really very close to the true local temperature.
Brad, it sounds to me (especially the counter-effect cooling had on the intensity of the emission) as if it was actually IR emission not adequately filtered in front of the CCD.
Did they verify the wavelength?
Wavebourn
Raising the temp of the whole opamp, will worsen the dynamic temp. unbalance.
A specific subcomponent when variably heated due to it’s I*R will find it more difficult to get rid of it’s thermal energy through the hotter surrounding mass.
The delta temp. with another subcomponent (less I*R loaded) will thus increase.
George
gpapag,
Thank you for the explanations George. I actually follow this quite well except my math is more than rusty... I am glad you pointed out the fallacy of increasing the temperature to control temperature as this just seemed completely incorrect to me. By doing that you have just shot yourself in the foot by what appears to be decreasing your chances of dissipating the rising die temperature to a lower stable temperature. Wouldn't you be better off using a direct contact heat exchanger, a finned heat sink than using air as the direct transfer function? I only suggested the thermocouple as they are so cheap and attached directly to the die with a simple hand held meter you would have a very instantaneous reading of temperature rise. You could also drill a hole as you stated and insert the thermocouple into that hole. I had not seen a hand held with that resolution of depth of field. By drilling the hole you would appear to have two affects, one being the properties of the cavity depth to width and the second being a much thinner substrate that has a time lag in response to the actual temperature change internal to the device. Am I following what you are saying properly? Thanks again.
Steven
Thank you for the explanations George. I actually follow this quite well except my math is more than rusty... I am glad you pointed out the fallacy of increasing the temperature to control temperature as this just seemed completely incorrect to me. By doing that you have just shot yourself in the foot by what appears to be decreasing your chances of dissipating the rising die temperature to a lower stable temperature. Wouldn't you be better off using a direct contact heat exchanger, a finned heat sink than using air as the direct transfer function? I only suggested the thermocouple as they are so cheap and attached directly to the die with a simple hand held meter you would have a very instantaneous reading of temperature rise. You could also drill a hole as you stated and insert the thermocouple into that hole. I had not seen a hand held with that resolution of depth of field. By drilling the hole you would appear to have two affects, one being the properties of the cavity depth to width and the second being a much thinner substrate that has a time lag in response to the actual temperature change internal to the device. Am I following what you are saying properly? Thanks again.
Steven
Actually, that is not the case.***
Interchip dynamics will not change unless the gradient to the outside world is different for each chip. If for example, one die's heat must go past the other to get out.
***the astericks are because the thermal conductivity of silicon is temperature dependent, the eq is:
Kt =286/(T -100) for T =300 to 600K
Cheers, jn
Hi Scott
I agree with everything you are saying here.
I think your circuit is the most interesting here so far.
I'll put it in to my simulator and have a closer look at it.
BTW: is this the model you are using for the BF862?
* BF862 SPICE MODEL MARCH 2007 NXP SEMICONDUCTORS
* ENVELOPE SOT23
* JBF862: 1, Drain, 2,Gate, 3,Source
Ld 1 4 L= 1.1nH
Ls 3 6 L= 1.25nH
Lg 2 5 L= 0.78nH
Rg 5 7 R= 0.535 Ohm
Cds 1 3 C= 0.0001pF
Cgs 2 3 C= 1.05pF
Cgd 1 2 C= 0.201pF
Co 4 6 C= 0.35092pF
JBF862 model parameters:
.model JBF862 NJF(Beta=47.800E-3 Betatce=-.5 Rd=.8 Rs=7.5000 Lambda=37.300E-3 Vto=-.57093
+ Vtotc=-2.0000E-3 Is=424.60E-12 Isr=2.995p N=1 Nr=2 Xti=3 Alpha=-1.0000E-3
+ Vk=59.97 Cgd=7.4002E-12 M=.6015 Pb=.5 Fc=.5 Cgs=8.2890E-12 Kf=87.5E-18
+ Af=1)
ENDS BF862
Cheers
Stein
Agree. I wouldn't be so enthusiastic about losing the cascodes, as besides higher unloaded gain they reduce the effect of thermals in the preceding stages. But a buffer is a good idea, then the compensation required won't be so load-dependent. Of course this is basic stuff.