Lumba Ogir said:
I underestimated speed of heating, thermal resistance and capacitance of devices used in current mirror, plus very deep feedback by current with own phase shift. Systems with feedback sometimes oscillate, the same happened with my working point stabilizer. Oscillations were very low in frequency (about ten seconds period), but enough in current amplitude to unsolder the resistor.
Claude Abraham said:
I did actually, this stuff is off topic but as I said your mileage seems to vary. As a standard product we have log amps based on the ancient Paterson diode connected transistor that span 10pA-1mA (8-9 decades) and with RE compensation are 10ths of a dB absolute accurate. You must have only needed relative log compression and no TC comp or maybe that was done somewhere else. Anyway it's OT.
john curl said:
If you bothered to read it states that the writers could not find log conformance below 100nA, they didn't look very hard. Modern bi-polar transistors are log conformant to 10pA and in some cases even lower. Barrie also has a book on log amps you might want to read up.
Claude - You have reduced this to such an abstraction removed from actual circuit design, it might be better to just offer an improvement on the art using this approach. Maybe a phono front end, that would be more on topic.
This is kind of sad, Claude is introducing us to his place in the industry. He has a patent and he is using it to show his credentials, and where he fits in engineering society. Whether it is relevant for the discussion, on hand, is secondary.
I have often found it necessary to do much the same thing when discussing engineering. One of the best ways to show this online is to point out a technical paper or patent, or even what you have worked on and who you have worked with. I have also been rudely rebuffed by others, when I did this, although they would bring out the organization that they worked for, and hint that they were surrounded by great people who gave them free advice and opinion.
Let's be polite in this context, rather than petty.
I have often found it necessary to do much the same thing when discussing engineering. One of the best ways to show this online is to point out a technical paper or patent, or even what you have worked on and who you have worked with. I have also been rudely rebuffed by others, when I did this, although they would bring out the organization that they worked for, and hint that they were surrounded by great people who gave them free advice and opinion.
Let's be polite in this context, rather than petty.
john curl said:
Nobody is debating this John. I guess Scott is curious (and count me in as well) as of why "a bjt is not as good a logging element as a diode". This is a very serious statement and some details on what was actually tried (and any available explanations) would certainly add some extra credibility.
I don't know what transistors do you mean, but back in 70'th I used a b-c junction to convert a voltage in a musical synthesizer. There was an old and dirty 4-transistor die, so I had also a heater and a temperature sensor in the same bottle. The instrument was always perfectly in tune, through all octaves.
Also, we used b-c junctions as diodes and b-e junctions as Zeners in thick film ICs because caseless diodes did not exist.
Also, we used b-c junctions as diodes and b-e junctions as Zeners in thick film ICs because caseless diodes did not exist.
john curl said:
Well stated John. I agree completely.
Scott, regarding the log amp, please consider the following. When scanning currency (vending machines, postal stamp vendors, car wash bill acceptors, video gaming machines) I had to determine if the bill was bogus or genuine. With color copiers, inkjet and laser printers, quality has improved to the point where distinguishing the real from the copy is much harder than it was a generation or more ago.
One of the methods employed is optical density measurement. This is well known in other industries as well. Light is transmitted from one side of the bill, and detected on the other side. Optical density, OD, is a logarithmic measurement. If the amount of light transmitted through the paper is 10%, the OD is 1.0. If only 1% is transmitted, the OD is 2.0.
Let's say a paper currency, or barcode coupon, or whatever is being analyzed has an OD that varies between OD 1.4 & OD 2.7. In percentage terms, a linear measurement would be very squashed, since OD 1.4 is just 4% transmission, and OD 2.7 is a mere 0.2% transmission. The entire signal varies between 0.2 and 4% of full scale. A log measurement makes the signal more usable. More a/d counts means better a/d resolution.
But, we need a logging element with at least 3 decades of log-linearity. In addition, the temperature dependence is very problematic.
If we examine the I-V characteristics of a p-n junction diode, we get the familiar Shockley diode equation:
Id = Is*exp((Vd/Vt)-1)), where Vt = nkT/q, you know the rest.
But we never force a voltage source across the diode in the forward direction. We force a current and then measure the voltage. Rearranging the familiar equation:
Vd = Vt*ln((Id/Is)+1).
This is a well known relation. But what is *seldom* ever mentioned is that "Is" the saturation current value, is extremely non-linear wrt temperature, increasing drastically w/ increasing temp. The Is value at -25 C is much lower than that at +25 C, and at 125 C it is orders of magnitude greater.
So if we plot V vs. I at any temp, we get a log-linear graph. But if we superimpose graphs taken at different temps, we get the following. Drive the diode w/ a constant current source, and measure voltage. At low temp, plot V vs. I, -25 C for example. Increase to 0 C and plot. Then plot for +25 C, +50 C, +75 C, +100 C. Observing the family of curves will exhibit the following.
As temp increases the voltage gets offset downward, but the slope increases. If we put 1.0 uA into the diode at -25 C, we get some V value. Increasing temp to 0 C w/ the same 1.0 uA will result in a V less than at -25 C. As temp increases the whole I-V graph gets offset *downward*.
But the slope *increases* as temp increases. At -25 C if we measure V at 1.0 uA, and then at 10 ua we get a larger V. Compute the difference. This is the delta V over 1 decade of current.
Now, increase the temp to 0 C, and repeat. The V value at 1.0 uA is less than at -25 C, and likewise for the V at 10 uA. But the *delta V* at 0 C is *greater* than that at -25 C.
Of course the diode is log-linear over many decades regardless of temp. But if we are scanning a paper note, when we read OD 1.7, then OD 2.5, then OD 2.2, what level of transmissivity does each value correspond to? That is the issue. It varies greatly with temp. In order to get reliable results, we must calibrate the amp/sensor system by reading OD 0 with nothing in the path. My servo circuitry adjusts a DAC or PWM to vary an LED current to get a brightness at OD 0 resulting in nearly the full a/d scale. At low temp, less light is needed to reach full scale because the V value is naturally higher for a given I. As temp increases, the amp output drops below full scale. This is sensed by the micro, and an LED is increased in brightness via higher drive current using PWM or DAC methods.
The detector amp then has a thermistor based op amp gain stage. The delta V increases in proportion to absolute temp (PTAT). So the op amp using fixed resistors and a thermistor varies the gain with temp so as to compensate the SLOPE characterisitics of the p-n junction diode.
To use a log amp under these conditions requires both OFFSET & SLOPE correction to null out temperature variations. Otherwise it won't work.
I've used practical circuit and application principles here with no advanced physics. The log amp works very well for scanning paper density and must be temperature compensated for offset and slope.
Then when a bill is inserted, the readings are taken. Does this help? BR.
They kept it at 85 degrees. Should be a military thingie...
syn08 said:
For one thing, the transistors I tested seemed to have a log-linearity region that spans about 5 decades of current. Bob Pease in his article about a decade ago ("What's all this logging stuff anyhow?") mentioned that a bjt exhibits 5 decades of log linearity. Bob and I have independently arrived at the same figure. For my work that was more than enough.
Another concern was that of V for a given I. We sought the largest V possible for a given I as well as the largest delta V for a decade of current. The bjt parts I tested were indeed good.
But the small signal diodes, specifically the 1N914B, and the JPAD series, exhibited more decades of log-linearity (around 8 to 10 for the JPAD, & around 5.5 for the 1N914B), as well as larger V and delta V. The larger V & delta V is desirable to get the most gain.
As far as decade span goes, the bjt had more than enough, but the temperature dependency favored the diodes. The bjt log-linearity was limited at the upper end by rbb', the base spreading resistance, and at the lower end by Ics, or Iceo, which is the leakage current of the reverse biased collector-base junction.
With temperature the Ics value goes up substantially with both devices, but the diode was not as bad. So 5 decades at room temp was diminished at elevated temp a little more with the bjt.
Without going into the atomic detail, I can estimate the reason using ordinary device properties. The lower limit is set by Ics, the reverse saturation current of the c-b junction. A bjt is optimized for signal gain purposes. The doping in each region greatly affects properties. Tradeoffs are made between Vce breakdown, beta, ft (speed) etc.
To get great log-linearity over many decades, what must be done? Obviously Ics must be reduced. The lowest value possible is needed. To do this, base doping density must be moderate to high. This also reduces rbb'. a good thing to do as well. The rbb' value limits the range at high current levels.
But, a high density of dopants in the base means that the beta value is greatly diminished. For a bjt employed in signal amplification, this is bad, especially for amps with a small number of stages/junctions. Most bjt devices are optimized for gain, not log-linearity.
A diode does not incur such tradeoffs. In a diode, the 2 regions, n and p, need not be optimized for any "current gain". The doping level need not be light because there is no beta to maintain.
The 2 regions form 1 junction. When forward biased, there is no leakage current present in a diode. In a bjt, with the b-e junction forward biased the c-b junction is reverse biased, and there is a leakage current from c to b. This leakage, Ics, combines with Ic, the logging current, producing an error.
Off the top of my head, that is what I learned when working with bjt and diode logging elements. Again, Vishay, I believe, has a "JPAD" family of diode parts optimized for logging. They claim over 8, up to 10 decades of log-linearity, a claim that I confirmed as valid.
A bjt will give around 5 decades, and incurs tradeoffs with rbb', Ics, vs. beta value. Since beta is not important in a logging bjt, the tradeoff works the wrong way. Maybe there are special bjt's optimized for logging use. But frankly, the JPAD diodes are really hard to beat.
Just my $0.02. Cheers.
![url]](/community/proxy.php?image=http%3A%2F%2F%5Burl%5Dhttp%3A%2F%2Fhome.comcast.net%2F%7Emnjmiller%2Fmosamp.gif%5B%2Furl%5D&hash=4d76350f74a8af4b2eed106a22c0ce6a)
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.