From: jl@glen--canyon.com
On Fri, 30 Jan 2026 04:24:18 -0000 (UTC), Phil Hobbs
wrote:
>john larkin wrote:
>> On Thu, 29 Jan 2026 17:22:39 -0500, Phil Hobbs
>> wrote:
>>
>>> On 2026-01-29 16:40, bitrex wrote:
>>>> On 1/29/2026 4:24 PM, Phil Hobbs wrote:
>>>>> On 2026-01-29 15:32, Liz Tuddenham wrote:
>>>>>> Phil Hobbs wrote:
>>>>>>
>>>>>>> Hi, all,
>>>>>>>
>>>>>>> I'm doing a high-accuracy version of the laser noise canceller
>>>>>>> .
>>>>>>>
>>>>>>> In particular, to get better cancellation accuracy, I want to
>>>>>>> get rid of the input offset voltages of a couple of op amps.
>>>>>>>
>>>>>>> One approach to this is to use a chopamp integrator to snoop
>>>>>>> the summing junction, and dork the noninverting input to force
>>>>>>> the summing junction to average 0.00000V.
>>>>>>>
>>>>>>> This is nice conceptually, but there are a couple of worries:
>>>>>>>
>>>>>>> 1. Chopamps kick out nasty switching spikes, which will have to
>>>>>>> be decoupled sufficiently well.
>>>>>>>
>>>>>>> 2. Weird-ass composite amplifiers always have weird settling
>>>>>>> behavior.
>>>>>>>
>>>>>>> I haven't done this lately, but I'm thinking of a TLV2333.
>>>>>>>
>>>>>>> Any wisdom?
>>>>>>
>>>>>> At that level of accuracy, beware thermocouple effects.
>>>>>
>>>>> It's all on one board, and the power level is low, so that
>>>>> shouldn't be a huge issue, I don't think. Gradients on the board
>>>>> should be way under 1K in the quarter-inch or so separating the two
>>>>> amps. I'll certainly put the power buffer some distance away.
>>>>>
>>>>>> If you are compensating a slow drift in offset, chop slowly and
>>>>>> sinusoidally, then the 'spikes' will matter less.
>>>>>
>>>>> I'm not the one doing the chopping--the spikes come from the CMOS
>>>>> switches inside the chopamp.
>>>>>
>>>>> The noise canceller works by splitting a larger photocurrent using
>>>>> a BJT diff pair, and adjusting the split ratio until the current in
>>>>> one arm exactly cancels a smaller photocurrent derived from the
>>>>> same laser.
>>>>>
>>>>> There are various fine points, but because the diff pair is a
>>>>> highly linear current splitter, the fluctuations split the same as
>>>>> the DC, so by adjusting the DC to zero, one in principle obtains
>>>>> cancellation of the fluctuations at all frequencies. A slow servo
>>>>> loop lets you do AC- coupled measurements down at the shot noise
>>>>> even with noisy lasers.
>>>>>
>>>>> With a bit of math, you can use the delta V_BE of the diff pair to
>>>>> do the same thing inside the feedback loop bandwidth.
>>>>>
>>>>> An offset voltage in either the TIA or the integrating servo amp
>>>>> causes the cancellation to be in error by
>>>>>
>>>>> delta I = V_os / R_F.
>>>>>
>>>>> With a 5k ohm R_F, a millivolt of offset makes 200 nA of current
>>>>> imbalance. With a 100-uA photocurrent, that limits the
>>>>> cancellation performance to
>>>>>
>>>>> Amax = 20*log(100uA/500nA) = 54 dB.
>>>>>
>>>>> It's better than that at higher photocurrent, but I'm chasing an
>>>>> honest 70 dB with this box, so the offsets have to be down in the
>>>>> tens of microvolts at most.
>>>
>>>>
>>>> So is the idea to LPF the crap out of the summing junction voltage,
>>>> send to a chopper amp used as an integrator, and then LPF the crap of
>>>> the chopper amp output sent to the TIA amp non-inverting input?
>>>
>>> Right, except that the output doesn't need filtering, just a voltage
>>> divider.
>>>
>>>> Is it spikes going forward to the TIA non-inverting input or going
>>>> backwards to the summing junction itself that's the most concern?
>>>
>>> The output is just ordinary noisy--53 nV in 1 Hz. Chopamp inputs kick
>>> out evil microamp-level spikes of low duty cycle--the 70 pA bias current
>>> spec is basically the bits of the spikes that don't average to zero.
>>>>
>>>> The non-inverting input is the devil and I don't really like it
>>>> anywhere but bolted to ground in precision applications but I guess
>>>> there aren't a lot of other places to inject a correction that isn't
>>>> going to disturb the summing junction worse
>>>>
>>>
>>> Well, it's got a 10k:50R voltage divider to help keep it still. The
>>> total adjustment range is thus +-12 mV or so, comfortably larger than
>>> the +-5 mV max offset over temperature.
>>>
>>> The voltage divider reduces the loop bandwidth by the same factor of 200
>>> for a given time constant, so to get a 5-Hz snoop loop bandwidth, it
>>> needs a time constant of
>>>
>>> tau = 1/200 / (2 pi * 5 Hz) = 160 us
>>>
>>> so the integrator has 200k * 820 pF.
>>>
>>> Since the 820 pF is connected between the inverting input and the
>>> (low-Z) output, it'll suck in most of the spikies, but just in case, I'm
>>> splitting the 200k in half and bypassing the midpoint with 1 nF to ground.
>>>
>> Couldn't the offset servo loop be mega-slow? I think it's correcting
>> thermals.
>
>I think several hertz is the sweet spot, because there are situations such
>as turn-on where it’ll start way out of whack and have to recover.
>
>It’s more the fixed part of the offset than the drift that’s the problem,
>except at extreme temperatures.
>
>Cheers
>
>Phil Hobbs
Trimpot!
John Larkin
Highland Tech Glen Canyon Design Center
Lunatic Fringe Electronics
--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)
|
