This LM3900 circuit is a building block intended for other signal processing building blocks. It's a paraphase amplifier, which in this case allows arbitrary inputs to be differentially centered about ½Vcc. Alternatively, another more precise reference voltage could be used as the "platform" for offset. A working example is the temperature stable 1N5232B Zener diode voltage reference circuit with LM3900, described in other posts. In the unipolar signal processing world with LM3900, I call such offsets Vertical Zero Reference or VZR. Paraphase outputs are useful for other circuits that require (or could benefit) from working with differential signals.
Paraphase Ampilifier
The circuit was a LM3900 adaption of similar circuit from National's LM359 datasheet for a balanced line driver. The LM359 is a broadbanded version of the LM3900, and still available -- albeit only in SMT form.
Compared to the LM3900 however, the LM359 has significantly stiffer output impedance and current drive. My LM3900 circuit cannot actually work (as designed) as a balanced line driver (e.g. with a common 600Ω load), because the 8kΩ output impedance is too low. I discovered that when attempting to drive a 604Ω 1% load differentially, the +drive output actually impinges on the -drive output driver -- distorting the waveform on just that side.
While fortuitous learning, a balanced load was not really my application. Instead, I wanted to drive two different matched loads, but with precise differential phase and voltage offset. These circuits would generally also work at higher impedance levels, hence the dual 10.0 kΩ 1% loads depicted in the notebook circuit above.
There were several other interesting insights gained from this circuit development. These insights have to do with what's required to achieve balance. Additional insights are around what the fundamental hard limits are with the LM3900 as a system building block.
As it happens, two cascaded LM3900 gain blocks, a +1.0 one followed by a -1.0 one, do track quite well together using 1% gain-setting resistors. The +1.0 gain block sets the commom mode offset (e.g. VZR) to ≅Vcc/2 using resistor R2 as 200kΩ 1% tied to +15V. This is inexactly Vcc/2 due to the +Vbe at the -input; even that could be compensated out, but it's unnecessary here, as the intended output signal swing is no more than ±5V, or 10Vpp. The measured VZR of 7.68V is good enough. The -1.0 gain block then matches this 1∶1, because first it's gain matches closely, but the VZR is also set to the same level. This is done using a 100kΩ 1% tied to +15V into the +input; this resistor sets the same level for the inverting gain block because it compensates it's own VZR, as well as the VZR output by the non-inverting gain block. The circuit matches well even untrimmed -- within ±20mV! My application needed precise matching, so a trim circuit for precise DC balance was added. This is the 100kΩ + 10kΩ cermet added to pin 13 for the inverting op amp (see engineering notes, above; and next). Note also that using two +inputs for a differential circuit has an inherent matching due to semiconductor device fabrication.
Interestingly, one sees "one-sided" sort of things in the unipolar LM3900 world. To get an adjustment range of maybe ±5%, to cover all corners, a cermet trimmer of 10kΩ is added to the 100kΩ VZR setting resistor for the inverting op amp. Instead of taking ≅5% less off the 100kΩ resistor, then adding ½ the trimmer resistance. The correct form of the trim allows the inverting op-amp to match whatever voltage level was set for the quiescent VZR voltage of the non-inverting op amp. The trim is on the inverting side, not both, but allows ± movement about the fixed VZR of the non-inverting op amp. This is because trimming the non-inverting op amp upsets the balance seen by the inverting op amp, which is supposed to precisely mirror the other side.
The trimmed paraphase amplifier works exceptionally well, providing clean 5Vpp output even out to 20 KHz under sinusoidal excitation:
Dual ±2.5V output into twin 10KΩ load at 20 Khz, ≅7.5 Vos
The LM3900 slew rate limits are evident though when attempting 10Vpp:
Dual ±5V output into twin 10KΩ is a brige too far at 20 Khz
The LM3900 pulse response at 2.5Vpp is quite serviceable at 10 & 20 KHz:
Dual 2.5Vpp pulse output
Typical LM3900 pulse response for 2.5Vpp at 20 KHz
This paraphase amplifier circuit design is AC coupled at the input. With the values chosen, the lower frequency -3dB corner is 0.159 Hz. This is low enough that sharp-edged signals at low frequency are still faithfully reproduced differentially at the output:
20 Hz pulse response
20 Hz sawtooth wave response
Generally, obtaining antiphase balance is a non-issue with bipolar op amps. But with the unipolar LM3900, the current mirror at the +input can actually be relied upon for good DC accuracy over certain ranges of current. Empirically, it seems to be that current range is ≤ 250µA. And there's a derivative insight from looking at 250µA input currents. If you look at the input bias current at the -input, Ib ≈ 30nA as the bottom limit, then the approximately largest available current range implies a typical dynamic range for LM3900 circuits:
250µA ∶30nA, or 78.4 dB.
You can get +6dB more, or 84.4 dB net, if you can allocate 500µA instead. Even more current is possible, but more deviations from linearity will occur. Around this point and beyond, you are hitting the maximum input current limit over the full operating temperature range of the device.
A system performance reference mark for the LM3900 is that it has a useful dynamic range of ≅ 80 dB. With care, this component can have good DC accuracy, but it's certainly not an OP177. The LM3900 has good AC gain for a general purpose device, with unity gain crossover at 2.5 MHz. But it's of course not the fastest nor lowest-noise op amp one can obtain.
Instead, unipolar circuit design with the LM3900 is about obtaining good results with a solidly reliable yet inexpensive component, esp. with application flexibility offered by its unique mode of operation. The circuit designs in this blog are to intended to support an electronic music synthesizer having new sound. These are not intended to be audiophile circuit designs, absolute performance is a non-goal. To paraphrase: "The enemy of good is better."
Some other important things were learned from developing this circuit. The paraphase application could have been done alternatively by cascading two inverting -1.0 gain blocks. In fact, just moving resistor R1 from pin 1 to pin 6, for op amp 1, directly enables this. But in LM3900 circuits, it must be remembered that the +Vbe present at the -input has temperature sensitivity. When two -1.0 gain blocks are cascaded, with zero input, the common mode voltage between them diverges under heat soak. However, when using a +1.0 then -1.0 gain block cascade, the common mode voltage does change slightly under heat soak, but it precisely tracks across both outputs! The +1/-1 gains are otherwise constant. The downstream circuit to use the differential output will see any temperature effects from this block as merely a small shift in common mode voltage at very low frequency -- which should easily be rejected differentially.
These engineering notes are bit more jumbly, unfortunately. I'll eventually clean things up with schematic capture. I'm still test-driving various tools right now, and they all annoy me. When I can decide on the least-worse among them, I can improve schematic cleanliness for some of this work. For now, however, hand-drawing is fast and accurate enough.
Comments
Post a Comment