Hot Transistors

I think that little can be done to fully stabilise the gain of an amplifier using only a single transistor, the technique is to design an amplifier with several transistors that has considerably more gain than is required and then reduce the gain to the required value with a negative feedback network.

The overall gain of the amplifier is then determined by the resistors in the feedback network that are much more stable than the transistors.

Temperature sensitive resistors can be incorporated in the feedback network to produce the reverse effect to compensate for any residual gain changes

Most of the earlier electronics was developed was discrete. Some how they managed this high temperature stability. Then there were these dual transistors in single casing with almost identical response. Perhaps, now ICs have all that inside. Sensing temperature and then correcting will be only good if they are in one pack. Temperature change of few degrees is very easy on the boards as now we pack a lot of electronics in small module. Transistor base emitter shifts with almost 2mV/C. Measurement of Microvolts shifts a lot. Some people used to mechanically chop the signal at about 100Hz to get rid of drift and then reconvert to signal below 10Hz. My signal is high frequency, but looks that gain is also changing for AC. Discrete are less noisy than all best ICs in the world I have tested.

NTC and PTC thermisters I have tried and have also filled the amplifier cavity with Alumina powder to make it oven like. I could get a bit better stability but still there is lots of gain drift and noise injected by the thermisters. I normally use 1ppm/C resistors and 1ppm NPO or COG special caps.

Chopper stabilised OP-AMP's are still in use but a frequency response up to 100MHz would not be possible, with modern transistors feed back amplifiers can readlly be constructed up to these frequencies while there may even be OP-AMP's available.

Twin transitors in one can are useful for constructing quasi logarithmic amplifiers but I guess its all done in a digtal manner today.

I bought these 1.5GHz OpAmps from Taxs and also from ADI. They do not give good output. However simple JFET along with high gain transistors works much better. So I use transistors now. I do not trust all those big numbers in OP Amps. They good only for high signal levels and 12-bit range.

Har HAr, a "hot transistor" is engineering slang for one that will oscillate at a high frequency, Ft

I hope you're not depending on the transistor's beta to set the overall gain figure. Are you?

What about an earlier poster's suggestion to incorporate feedback and set your loop gain by other means?

What must be your amp's gain/bandwidth product? What's it's 1/f. It's rolloff frequency? What it's noise figure?

To get exactly what you want, you sometimes you have to bite the bullent and build your circuit out of (mostly) discretes. Did someone tell you there's something wrong with that?

After you've done all that can be done, as a last resort you might think about using a well-regulated oven. Like a crystal oven.

Yes, I do use a feedback scheme for gain control for Dc and AC. However that is only a fraction of the output injected in the input. This affects the gain a bit. Yes, I did design an oven type thing with high damping factir such that temperature inside can't fluctuate rapidly. Perhaps I will now fill the gap with Polyurethane. Getting temperature control better than 0.1C is very difficult. I can do some gain adjustment using JFET. This is not taking me to +/-1ppm/C gain stability demanded by the user. Perhaps +/-10ppm/C is possible. I usually was wrking at +/-100ppm/C gain stability and now have to move two decade down. Some people use differential design and pack two transistors in one pack. I am using non-differential structure, which makes it difficult to have mirror current control. Perhaps I can have one reference amplifier and one working amplifier and then compansate for the active gain control. This will become a big circuit and may inject more noise. I am working with 1fC level charge detection and hence already very close to the noise limit. Heating will sure kill the amplifier. I can restrict the working environmental temperature within 20C to 30C at best and typical 25C+/-5C. This what one will get in an AC room.

Hy Shyam

Try Peltier Plate Heat Sinks; you can stabilize interactively the temperature of almost anything with these plates, you can convert this variable in a constant of the transistor model equations. To sense the temperature you can use an electrically issolated thermocouple attached to the transistor case.

As temperature is a slow varying variable, you must make sure the Laplace model is stable by slowing down de control speed. See this web page of a major manufacturer of active temperature controllers.

http://www.tetech.com/

However, on trying to control temperature you must invest energy, and this is an unavoidable law of Mother Nature. So first you must run your amplifier without temperature control, and them decide a temperature just slightly below equilibrium, other wise you will waste a lot of energy, the idea is not to reach a cold temperature, just to reach some stable temperature to eliminate it as a variable of the equation by turning it as a constant.

Jaime Soto

http://www.matharts.cl/

Good Idea. I will try that.

Well if the heat is what you want, you could use the method that was/is used to stabilize crystals used as time bases. Heat the transistor to a fairly high temperature and regulate the heater to maintain that temperature. As the transistor heats the external heater ramps back to maintain temperature set-point. It is a simple solution that is not invasive to your circuit.

I think the point is being missed here.... its all very well to have complicated ovens and multi-stage amplifiers with feedback etc...

But the actual problem is in the circuit design... A few tweaks and modifications and you wont have this change of ac gain with temperature...

I'm surprised you DO have a problem!!

As for using resistors and capacitors with only 1 ppm per degree centigrade temperature coefficient - That is plain lunacy! Not only must they be costing an arm and a leg, but they wont help!

John.

I am still puzzled why 'you're people are asking for a gain stability of 10ppm'...... Why???

Its only a pulse amplifier... its not a critical linear amplifier... so why have a ridiculous gain stability of 10 ppm!!??

CRAZY!!

John.

Dear John

Signal has 20-bit definition. That is why all electronics is becoming difficult. I have already reached 16-bit accuracy but going up further is becoming a problem.

Perhaps injection of tracer pulse and monitoring it for gain correction looks one option. This was earlier tried using radioactive tracer in the scintillators with peak at one point and gain was corrected using the location of that peak. Now we are entering into greater challenge in design of pulsed charge amplifiers that can provide 20-bit analog peak height definition. Now every part is picked up for that definition. I am also using temperature stabilized 24-bit ADC. Problem is still with amplifier and peak detector. I did receive some help from Intersil and they have faxed me a good design of the peak detector circuit using their parts. Unfortunately they have not given me permission to display it so can't place it on forum.

I am designing what no one has reached so far in this area in the form of commercial product. You will not get 20-bit nuclear ADC from any commercial source. I am just designing that. I can pay US$10000 if any one can make one for me. I thank you all for sincere comments. I am not sure how best I can design, but improvements are sure possible and one day I will be there.

I wil welcome all 0.1ppm/C parts for this design that have minimum 3 years stability. Some of these I already got developed. Now I am concentrating on active parts.

Heat the trsnsistor and regulated the temperature or measure the temperature and regulate the gain, which will be more stable from your point of view. Measurement of temperature and gain control may be with greater error.

Temperature compensation with temperature control is another option. But can we have 1ppm stable temperature compensation?

I would dearly love to know where you can get the 1 ppm per degree stable capacitors and resistors from??

A resistor maybe just about but not with age, a few months or a few volts and it will have drifted by more than 1 ppm...

As for capacitors, the best I've seen are about 1 ppm for a few months, but they are specialist calibration devices made from fused silica AND kept in a high accuracy oven, even then they have a ageing of 2 to 3 ppm per year...

Oh and they cost about $6000 each!

So if you're using capacitors and resistors with tempco 's of 1 ppm please tell me where I can buy them?

John.

Capacitors are from Vishay USA www.vishay.com/ web link. They make special 1ppm/C chips. I get these from 1pF to 100pF but higher values may be possible. They also provide highly stable resistors. I think these are silicon chips and not NPO or COG.

Others provide 30ppm/C COG/NPO caps and I have up to 0.1uF value. I do not pefer X7R if I can get hold of COG / NPO. I have million capacitors of this type in my store now and almost all values I can get hold of. I usually prefer high voltage version due to low leakage from them.

I even have Teflon capacitors for the lowest possible leakage. They are 0.1uF 400V DC/AC. These are very very expensive.

Hmmmm I think you're confusing 1 ppm tempco with the tolerance of 1%...

I certainly can't find any less than 1 ppm tempco capacitors from Vishay or anyone else...

If you are capable of getting these capacitors please let me know the part number as they should cost in excess of £500 each.

Maybe you have seen the material charactoristic of npo grade as being 0 ppm per degree C...?

and forgotten about the tolerance of this being +/- 30 ppm - Never mind the aging effects...

To be honest, I think you are being led astray by the customer overspecifying the requirement, this is very common in the nuclear industry. I have often been asked for ridiculous specifications from them and have designed equipment that does the job and is a realistic specification for what they needed it for.

But please, Shyam, this is a professional engineer's forum, don't think you can invent a new type of ultra high spec. component to dazzle us by when nobody else in the world knows about it...!!

All this reminds me of the Sci Fi film where engineers get an alien component catalogue and build themselves an interrositor or something!!

John.

Dear John

The capacitors given by Vishay are actually <1ppm/C TC. These caps are not NPO by semiconductor chips. That is what they told me and I used these in ultrastable charge amplifier designs for 1fC level measurement. Perhaps they might have discontinued these now. These parts were developed for me 3 years ago and they took few months to get these developed as they were initially doing only 1pF and 2.2pF range and I wanted specially 100pF. Vishay has become very big now and they may be doing some restructuring.

NPO caps are pretty good but are not specified for better than 30ppm/C. They do have high stability points and that is why I use this oven in amplifiers.

I am not talking about initial tolerance as that can be numerically corrected. I am talking about drift in operating temperature range. For example best stability in quartz crystal is +/-1ppb per year drift in oven controlled devices. We have no other material with better LCR properties.

I am going to create some standard setup for primary level measurement (actually secondary NIST traceability). http://ts.nist.gov/traceability/

I need to do long term measurement in ovens for years, online or in switched state to see if material drifts. Most of the material I use is for almost the limit of the specifications. Just some measurement is not enough. Reliability of specifications is equally important and I need to send samples in labs that maintain primary standards.

Even for 24-bit ADC the big question was to trust the DAC or ADC for error computation. Assume that you digitize the input and again make an analog and take difference then error is assigned to ADC or DAC? There is really no other way arround.

Perhaps it will be great thing if we discuss on primary standard for few things and way to create such facilities. Are you interested in joing this? I am puting this in new technology link as separate idea.

Can you tell us the official name of those capacitors with 1ppm per degree for the thermal drift because I can't find this via vishay web-site.

If you bought it then you can easily find the name of those components. And I can download datasheets for those components, check the required characteristics and order it in the future for ultra-stable applications.

Thank you very much in advance.

I think these were made from silicon chips and are in 1pF and 100pF range. It is likely they may not be produced now. These are in tape reel that I received from them. They are not NPO or COG. Perhaps were made from silicon oxide which is also used as floating gate in EEPROM. They are white color chips of 1mm.

You can also get such capacitors for limited temperature range with high stability.

It's very strange to see the answer like this. I specified my question enough to be sure that answer would be enough clear for me. I didn't ask about dielectric type and color of chips. For me it's very important information - name of the components (name given by company). Of course, CAPACITOR - is not name, also like value, dielectric, voltage and etc. Also it's very strange to see the size of this ultra-precise components. With value of (1-100)pF and size 1mm as you said - we have some problems with the term and thermal drift of this components when we solder this onto the PCB (stray capacitance, leakage current and etc). 1ppm per degree means in absolute value for the 1pF capacitor something like 1aF per degree.

Chronoman, I think you will find that there is no practical capacitor that exhibits less than 1ppm per degree temperature stability, or even close to that... unless it is specified as 0 ppm +/- 30 ppm etc...

The best laboratory capacitors used for calibration only possess this stability due to them being housed in precise temperature controlled ovens... and they are the size of a shoe box costing many £1000's...

John.

I agree with John that it is very hard to have stable capacitor. Some temperature sensing and trimming may have an average effect on capacitor value. Often oscillators are made in this way. Best possible arrangement I know was 10^-9 or 1ppb quartz crystal oscillator.

If you place few kg brass in a thermus flask and then it will find the temperature around it stable for bery long time. Thermal inertia and isolation from external world will make the metal at stable temperature. Unless one is generating lots of heat, this works well.

To design the zero temperature ciefficuent capacitor, you can take NPO and then cut it in such a way that A/L ratio is same for all temperature to make a fixed capacitor considering that dielectric does not change with temperature else that also to be compansated.

Control all right but how will you measure it. Some spectroscopic transistions may be very sharp to temperature to milli degree C. Most of the material that are subjected to strain of temperature or physical strain will show temperature drft due to inner molecular rearrangement. Hence, one have to keep the material at a particular temperature for months to get a stable property. It is like stabilizing the atomic structure. It will be much better if this is done at very low temperatures near liquid Helium temperature.

Okey, question's gone. I just asked about it because I saw this information about 1ppm per degree for the capacitor. And I made for the my current project some comparative analysis for the different reference components (resistors, capacitors, voltage references, quatz oscillators and so on). and for the best of the capacitors I used 30ppm per degree. For the my current design (capacitive displacement sensor with 1pm resolution and 20-24 bit dynamical range) the capacitive reference with 1ppm per degree stability looks very nice (instead of 30ppm). But the actual state of things give us only 30ppm. Sorry for the opening of that question.

Maxim makes 1ppm stable oscillator with capacitive tunning and chip can sense capacitor change with temperature. ps jitter stability is possible. Squid sensor have 7 decade dynamic range and the tunning capacitor is very stable there to 1ppm level. It all depends to what you are working at.

Assuming that we offer you a capacitor of 1ppm/C stability from say 24C to 26C or tell me what is you range, then are you willing to pay for it? I think I will like to do some research on this if there is good money coming for the work.

Stability has temperature span and period for which it is stable as drift is assured. There is vapor pressure so even atoms evaporate or migrate, diffuse, get elected and Fermi level is not a sharp line

Thermal noise can be computed and this is something you have to live with.

Let us say Pounds 1000 for each capacitor 1nF of 1ppm/C per day drift for 24C to 26C range, will that satisfy your need? Let us get to the point of commercial feasibility imagination if not a real order. I think Jon will be serious only if there is some money in it and that is true with me too.

Not even the highest technology incubators for early birth babies have 1 ppm precision, I have seen incubators with complex CPUs connected via Internet to the doctors studios, but even with all that sophistication they never attain 1 ppm temperature precision, that is impossible, and I guess you don't need that precision.

The other path is to control gain from a high gain Opamp used as a preamplifier, that is classic analog design, you may simulate it first using MicroCap 8 or similar. In this case you must first take a sample of the output signal attenuated with a directional coupler, and feedback to the OpAmp, provided the OpAmp has enough bandwidth to handle these signals.

In relation to 20-bit resolution, my philosophy is to see it from the noise floor stand point. I mean, that it is not important to have "DC 20 bits resolution", but "AC 20-bit resolution". If a solution needs to have DC 20-bit resolution, then something is wrong, it is like trying to measure the distance between two cities in millimetres, just hire 50 persons to measure that distance with the best technologies, and you will get 50 different samples, with errors in meters.

Jaime Soto Figueroa

http://www.matharts.cl/

ADC we talk only differential and integral non-linearity. Here drift with temperature is a problem for gain and offset both . We are not talking about noise right now. We have signal to noise ration of 20-bit or one out of one million or better. Right now ADC is 100KSPS and will take it to greater rate using parallel array. Signal is fast and random so I detect peak and then degitize. If a pulse is missed then I use a computation for statistical correction. There is 10us dead time. I will like to reduce that to 1us in future and evaluating some fast mixed ADC technique that was developed by inhouse. It uses less accurate fast ADC and then corrects the data accurately using a scheme. Ortec also tried some correction technique but our technique is far better and proven now.

Early Video tape recorders used a technique of FM for the relatively low frequency components of the signal up to 100 kHz and direct analogue for the high frequency components.

Could you use something similar with a chopper stablised amplifier for the low frequency components

Yes, I have used those multi-channel recorders that could go up to 500kHz. They have poor dynamic range for high frequency sugnals and tape recorders were never meant for greater than 50db dynamic range. Only some audio recorders were able to reach 60db. 80db must be extreme limit in audio systems.

At about 70mA to 100mA collector current the gain of the transistor almost becomes least dependent on temperature. However, at such a hgh current, input signal need to be very high and gain is almost near 50.

Perhaps for some transistors this may be at lower collector current.

Some one will have to work on it, but you can fix a NTC thermistor on the hot surface of transister and connect between base and ground; It will work as negative feedback system.

This is for NPN Transistor, and one can use PTC for PNP Transistor in series with the base resistance,

There is one big problem with thermisters is they are unstable devices and over period of time they drift in value much more than PT1000 sensor. I am using Platinum resistance thin film ink for sensing. Problem then becomes that film senses faster than chip inside the transistor. Perhaps a transistor created in the same transistor semi will be a much better idea. Hence, will try dual transistor device. I have used LM199A reference which has a heater.

This voltage reference LM199AH actually works with high accuracy. Guaranteed 0.0001%/°C temperature coefficient. I need something like that for a transistor gain. It may be about US$30-50 range. I have MIL Grade.

I think that LTZ1000 has better temp. coeff. (0.05ppm/K) and long-term stability with the same costs but of course there are some advantages and disadvantages there.

Dear Chronoman

LTZ1000 is a voltage reference. I am not sure if this was intended to be used as amplifier . The two transistor inside are low frequency DC level amplifier for the diode reference, which is at heater temperature. This is similar to national semiconductor LM199 reference diode, which did not have those transistors.

Have you used the transistors as amplifiers? Can you be more specific to your point here?