PI Project update

[Anonymous]

OP 27 is internally compensated. OP 37 is decompensated and requires a closed-loop gain greater than 5 up in the megahertzes. A capacitor in the feedback loop more than a few PF will normally be enough to cause ringing or oscillation.
The apps circuits which show feedback capacitors are circuits for the OP 27.
Both are good op amps pretty much free of rude surprises, which is why they're still popular after all these years.
--DJ

 

[Anonymous]

Hi Dave,
Should probably let this go but the datasheets in the 1995 Burr Brown data book show applications of the OPA37 (not 27) for a low noise RIAA preamplifier with a 0.01uf capacitor in parallel with a 7.87k resistor, and that in series with a 0.03uf capacitor, all that in parallel with a 97.6 k resistor, in the feedback loop. So the two capacitors are in series from the output of the amp to the negative input.
Another app for a NAB tape head preamplifier (again opa37) has a 0.01uf in parallel with a 316k resistor in the feedback loop.
Input lag/lead compensation is shown for unity gain inverting amplifier which is probably better to preserve bandwidth. Of course for gains less than five the 27 is probably a better choice. But these gains of 500 to 2000 more than satisfy the stability criteria for most these decompensated devices. Then add some cap to reduce bandwidth to what is needed in order to reduce noise.
Thanks for the feedback.
JC

 

[Anonymous]

This is the other amp. that I have used successfully in front ends, although mainly in industrial detectors. Originally an RCA product I believe. I found that this needed tweaking with a small cap. across the feedback R for some applications. 1pf was too much, so I used two short lengths of wire wrap wire across the FB resistor and twisted them together till I got the best response. i.e. quick recovery with no overshoot.
Eric.

 

[Anonymous]

Tell us how you really feel, Dave!
IMO, there are 3 reasons data sheets are the way they are:
1: Companies don't want to advertise bad specs.
2: Ignorance of applications.
3: Historical, i.e., that's the way it's always been done.
Number 1 will get exposed pretty quickly in certain applications, but in many applications those same specs may not make any difference. Yet a system designer who doesn't know better might still choose a part based solely on a comparison of data sheets, where the major difference was a spec that didn't matter. So the part might get spec'd for the dominant applications, where it is fair to ignore a certain parameter.
Number 2 happens when you're unaware that a spec is important to customers. I joined Analog Devices when we released our very first CFA's (designed here in NC!), in response to Comlinear. Not fully aware of all the nuances of CFA, we spec'd them pretty much like ordinary voltage feedback opamps. In a quick look at our AD8000-series CFA's, I see we are now spec'ing things like inverting input current noise. So things have progressed.
Number 3 is an ongoing issue as markets develop beyond the traditional scope of a product. I design high-speed ADC's which have found wide usage in the telecom market. We still spec SNR, SFDR, jitter, and other traditional ADC parameters. But what our telecom customers really want is noise figure, input IP3, -1dB compression point, and other radio-parameters. The problem with historical baggage is, what company will be the first to step out of that box? Chip companies tend to follow each other on data sheets simply because they know customers do data sheet comparisons. If you step out with a radically different data sheet, even though it's the right thing to do, you might confuse the customer and lose a socket. Inertia that's hard to overcome.
Regarding comparators, I personally can't imagine why a company would not spec input-referred noise (either in rms noise or an equivalent jitter), yet in looking at our parts we do not. When I design a high-speed ADC, input clock noise up to the first sample-and-hold is a key performance limiter, as jitter directly degrades SNR at high frequency. So I spend considerable time designing ultra-low-noise clock buffers (comparators). The 14-bit converter we just released has less than 100fs (yes, <I>femto</I>) of rms jitter!
I will try to ask the comparator gurus why this is not. Personally, I think all chip designers should spend at least 2 years in apps to learn what is and is not important. I did not, wished I had, but I do customer visits to try and keep on top of things. One of the reasons I jumped into metal detectors was that, in pure chip design, I never get to actually build anything and discover these quirks myself, and I never get to hold a usable end product in my hands. All I have are big color chip plots hanging on the wall.
- Carl

 

[Anonymous]

Carl:
I realise this doesn't make it easier on the manufacturer, but....
For products that have two or more radically different apps (in terms of what the customer needs to know), publish two or more different data sheets.
An obvious possibility would be for op amps, publish one data sheet for the traditional linear applications, and another data sheet for driving A/D converters. Each data sheet would have just enough of the whole picture that it could be compared with other manufacturers' data sheets who are trying to cram everything onto one data sheet and not doing a very good job of it.
I occasionally see apps which use op amps as saturating comparators (for instance because they need the low noise, or the low offset, or because it's a leftover section in a dual or quad), and which use op amps in circuits where they can saturate. Both things are common in metal detectors. So.... how about some information on dynamic saturation behavior?
OK, that leads to another possibility. Right now we often have short form (usually single-page) data sheets used primarily for comparison shopping, and complete data sheets (typ. 6-35 pages) for design purposes. For popular and specialized products, there are often separate app notes.
I propose adding to the pile, perhaps available only for download, a third-level data sheet, perhaps twice the size of the regular data sheet, with additional nonstandard information furnished in a nonstandard way, and not part of the specification. Just a pile of additional info from the engineering dept. for those who are using the op amp or whatever, in a critical app or in an unconventional app.
--Dave J.
--Dave J.

 

[Anonymous]

I was thinking about the inverting input noise current in a CFA... seems that it will be reflected through the equivalent parallel combination of the feedforward impedance (the coil network in a PI), the feedback impedance (very high), and the input impedance of the inverting input itself. Since the latter is usually 20-30 ohms, then a higher coil network impedance would not matter too much, and the inverting voltage noise is dominated by the current noise times the input impedance. Am I missing something?
- Carl

 

[Anonymous]

A comparison with the other circuits, as well as plain old ordinary analysis, would lead one to the conclusion that the RIAA preamp circuit they show as an OP 37, is a mistake, and that they intended it to be an OP 27.
The .01 and 100 ohm resistor (neglecting the .03 for simplicity) are 3 decibels at 159 kHz. According to the data sheet, the OP 37 should oscillate in this circuit at about 30 MHz.
--Dave J.

 

[Anonymous]

I can already see resistance to multiple data sheets:
"Do you have the data sheet for the AD1000?"
"Yup, but it doesn't mention NF."
"Which data sheet do you have?"
"Don't know, how many are there?"
"Well, we did a generic, one for ultrasound, and one for you radio jocks."
It could become confusing quickly. I would favor simply extending the data sheet to have separate tables for major application groups. And, as you mention, there are the app notes.
One thing you may have noticed, at least from ADI (I don't pay very close attention to other mfgs), is that our data sheets really are expanding to include way more info, especially in the back page plots. We now include all sorts of worst-case plots, and even show statistical distributions of some parameters. It's getting to where a simple opamp has a datasheet as thick as some magazines.
But I agree, we should include more application-specific data, and maybe even more esoteric data.
As our last Prez said, "I feel your pain."
- Carl

 

[Anonymous]

The problem with the inverting noise current is the voltage it drops across the feedback network.

 

[Anonymous]

Hi Dave,
Tried a bunch of feedback capacitors from 1pf to 0.47uf and the bandwidth kept going down, but no stability problems. Of course reaching unity gain crossover at a lower frequency so should be more stable not less.
I agree there can be mistakes on data sheets, sometimes the mistakes carring through various companies being bought and sold. I did see some mistakes corrected by the third company who bought another. In this case I believe Burr Brown (haven't noticed too many mistakes on their datasheets) meant the op37 in order to have the large gain and still have bandwidth left.
I remember back in 1977 I used to compensate the LF157, of the LF155,156 series with a parallel cap to set the bandwidth I wanted. Now compensating decompensated amplifiers below their stable gain is best handled by following whatever recommendations the manufacturer suggests as this can get strange.
JC
JC

 

[Anonymous]

Yeah, but back in the bad old days, all manufacturers published complete schematics of their IC's. That's rapidly vanishing. I doubt that they're protecting IP, although that could occasionally be a concern. Rather I suspect that there are so few OEM engineers left who still understand IC guts, that the IC manufacturers figure why bother to furnish the information any more.
The widespread use of SPICE models has reduced the need for schematics and some data sheet parameters, but some of us old farts still design with a pencil. I realize however that the handwriting is on the wall, and the time is coming when I'll have to trust black box IC's and their simulations.
Makes me cringe to think of it, having to design stuff without the privilege of understanding it, but what the heck, I trust most digital stuff without worrying about what goes on in between the ports. And A/D converter mfr's rarely tell you anything about the internal comparators that decide whether a bit sinks or swims.
---SEPARATE SUBJECT: FLICKER NOISE
The conventional thinking in linear IC design is that in order to eliminate flicker noise, you gotta sample the output and then store a correction value, typically on a capacitor.
There is another way to do it I've never seen done, except in metal detectors where it's a by-product of half wave demodulation when using OTA's such as the LM13700 (an incredibly versatile building block that was taken out of production because so few OEM engineers understood IC's at the transistor level).
Here's my theory. Plain vanilla flicker noise is a current time-correlation phenomenon. Turn the current off, and you erase the time correlation. So for instance, if you want to eliminate flicker noise below 1 KHz, you switch the differential pair current on and off at a 1 KHz rate, and the deed has been done.
Now this method does not do anything for DC offsets, or for secondary sources of flicker such as thermal drift. And, it still leaves you with the necessity for switching. So, it is not of widespread applicability. But in mixers and switched amps (I have used OTA's as gateable preamps in PI's) the effect can be useful.
In communications chips nowadays, with so much stuff headed for quadrature demodulation direct to baseband, flicker noise is a real engineering challenge. I suggest that designing the downmixers to cut the current off will help reduce flicker noise.
Many years ago I passed this idea in front of Bob Pease, and he didn't know what to think of it, which kinda surprised me. It was at a convention, and his mind may well have been on other more important matters, like how to get yer gizzard to stop rebelling against all that greasy food ya gotta eat when you're on the road.
--Dave J.

 

[Anonymous]

If that's the one I think it is (sorry, I don't have an RCA linear catalog any more), it's very noisy, especially in the flicker spectrum. It was popular in its day because it would operate at low voltages, was fast, and would swing rail to rail.
RCA failed to say in their data sheet that the offset would drift up to several millivolts over time. I found out the hard way. It made worthless thousands of dollars' worth of solar radiation data. If they'd have warned about the secular drift, I could have just thrown in an offset to keep the output above zero, and the offset could have been subtracted out of the data later during data analysis......ANOTHER! example of the manufacturer thinking that if you conceal a serious limitation, the user won't notice that the product doesn't work.
--Dave J.

 

[Anonymous]

Hi Eric,
The Ca3130 was a RCA product, back in 1975 it was one of the fastest around, used it on a radar project for doppler (up to 100khz) processing.
Remember needed +-6volt supplies for the thing.
The twisted wire subpuff capacitor is called a "gimick" by the ham radio guys who end using alot of them.
I have a design for a 15khz pico amplifier with a 1 Gigaohm feedback resistor. End up needed a "gimick" to tune it just right, because of the pole formed by the capacitance of the resistor. There is a way of using a network in the feedback loop to allow the tuning with a pot instead of the "gimick". Really trick, and works great.
Thanks again for the info.
JC

 

[Anonymous]

But noise current, like other current, seeks the path of least resistance. Which, in this case, is the inverting input impedance (which is very low).
- Carl

 

[Anonymous]

Schematics aren't included anymore mostly because of their size - even a simple opamp these days can be 100 transistors. And the full schematics can be too much even for good transistor jocks. Nowadays, we put "equivalent" input & output circuits on the datasheets. I, too, prefer the good ol' days of looking at the full schematic. I keep a lot of antique data books just for that reason.
Flicker noise ---
Being a mostly bipolar designer (now on BiCMOS!) using some really hot-rod processes, I've had the luxury of not worrying too much about flicker noise. It's one of those MOSFET things.
My current design is actually an IF receiver for telecoms, and I'm taking a 300MHz IF directly to I/Q baseband. The LNA and mixer are bipolar, and any flicker that might exist will get mixed away anyhow. After the mixers are LPF's which have opamps with bipolar front-ends specifically to minimize flicker noise. By far, thermal+shot noise has been the bigger challenge.
In fact, I'm sitting at work, running SPICE sims on the VCO divider (at 2.5GHz!), trying to figure out how to generate phase-accurate I/Q clocks for the mixers.
- Carl

 

[Anonymous]

When you crunch the numbers, the resistance of the leg to ground in the feedback network, has to be extremely low, or the noise figure will be impaired.
In many designs this is no problem: either a low noise figure is not needed, or extremely low resistance presents no difficulty. However, in hand-held battery powered apparatus, and where CFA's are used as preamps, both considerations are important.
--Dave J.

 

[Anonymous]

Phase-accurate I/Q:
Good news: a ring counter (not ripple counter + XOR gate) is pretty good at this.
Bad news: it has to be clocked at 4x frequency.
PS: I'm used to thinking below 200 kHz. What is most practical at 2.5 GHz I don't really claim to know.
------------
Here's my suggestion, Carl: give up, throw in the towel, use a method that is attractive from every standpoint other than accuracy.
Then, cheat. Bugger the sucker with laser trimming, EPROM-controlled something-or-the-other, etc.
If a supe bellyaches about the cost, throw in another feature or two that also requires buggering, and remind him that the result is cheaper than continuing to try to fight the problem without doing what's necessary to solve the problem, and it advances the state of the art where you work, at the same time.
-------
Howzabout this. You use an analog phase shifter using a varactor (or possibly a variable resistance, or variable transconductance), to achieve quadrature. In production you hit it with a valid provoke signal (you're gonna do that anyway, right?). You use the unwanted image to servo an EPROM, which will probably be there for some other reason anyway. The EPROM controls the varactor voltage (or other variable element).
Or: The EPROM itself actually IS the active element which controls the phase shift. However, that kind of cleverness is probably unwarranted these days. Makes better economic sense to not be clever, and just do it the obvious non-clever way, so the design tools handle it well, and your fellow co-conspirators understand what's going on without any need for elaborate explanations.
---------
Meanwhile, back at the ranch where the John Deere was parked, on account of someone swiped the battery to run a beeper: is the physics of the current chopping scheme I proposed valid, and is there an app for it, possibly at low frequencies rather than high?
And: the only new OTA's I know of, are video speed current hogs. Is there any inclination in the industry to replace the LM3080 or 13700 with something having fairly similar topology, but improved processing? Better count the cost of good app notes in the production cost!-- most OEM engineers nowadays don't have the foggiest notion of the myriad things that can be done with OTA's.
Incentive plan and design criterea. It's a single OTA, bipolar, fairly similar to 3080, but improved processing. Since it's improved, and aimed at the savviest OEM engineers, it doesn't have to be designed to rock-bottom cost like the 3080 was. So, do nifty stuff to it. Make it speedy. Use super beta transistors on the input. Offer a control pin that throws emitter resistors in the current mirrors to reduce input-referred noise. Offer another pin that throws in emitter resistors and/or possibly cross-coupled transistors, to improve input signal-handling capability without compromising noise too much. Offer another pin that provides a temperature-compensated bias voltage or current, including a topology that allows a noise reduction capacitor to be added by the user. Trim the offset. Use a low noise process. And.... furnish an additional pin that makes it possible to couple a capacitively-coupled fast turnoff signal so flicker noise can be eliminated at moderately high frequencies.
Early effect presents a marketing dilemma. You gots your choice: tiny geometry, high speed, low output impedance; or big voltage, slower, high output impedance. Methinks that the market would be better for the slower and higher output impedance, but if one version would be profitable, the other could probably hitchike the profit trail as well.
Now: the secret agenda you don't have to tell your employer about: what you're really designing is a slice of silicon for PI front ends!
Watch this: we load the output with a couple of back to back diodes in parallel with a resistor, and guess what? it's fast and doesn't have to go into and out of saturation.
Watch this: we turn the sucker off momentarily during transmit in order to decorrelate the flicker.
Watch this: there is no concern for preamp feedback loop stability, because there is no feedback loop.
Watch this: since current turn-on is inherently fast, the current control input can be used to gate the receiver, possibly in conjunction with separate CMOS analog switching ahead of the preamp input.
Watch this: since the output impedance is high, the output current can be switched into an offset correction capacitor. This may not seem important until you start asking what happens at high DC gains and low voltages; or when you start asking what it may have to do with flicker mechanisms which are not cancelled by current chopping.
Enough for now. The walls of the box are not the limits of the Universe, but the benefits of evolutionary selection make it desirable to put tentative boundaries on things so choices can be made. Otherwise, the "observer" of Schroedinger's Cat Paradox, who can't figure out whether to obey Heisenberg or the Second Law, will be standing in line for public dole.
--Dave J.

 

[Anonymous]

it just occurred to me that the b-word I used in the previous post, which doesn't raise eyebrows much in the U.S.A., might be deemed more offensive by our non-Yank friends.
I enjoy using slang, to add levity, not to offend. Please accept my apology, and I'll try to remember not to use that word in the future on this forum.
--Dave J.

 

[Anonymous]

Carl if you looking for numbers and intrest count me in

 

[Anonymous]

Hi Andrew,
My personal opinion is that much high end audio technology is based on marketing hype rather than technically supportable facts. Having done an investigation into speaker cables in 1996, I came to the conclusion that many cheap cables are superior to those costing hundreds and sometimes thousands of dollars. In fact I use the same coax cable as on my metal detectors, both as a speaker cable and interconnects. The quality of sound I get with my system does not give me any desire to spend money on anything else.
Oxygen free copper has a higher conductivity than standard copper but it is not clear to me how this can make a difference in noise.
Eric.