On Tuesday, October 24, 2017 at 4:51:27 AM UTC-4, Steve Wilson wrote:> dagmargoodboat@yahoo.com wrote: > > >> The ringing period can be estimated as Period[ns] = 2/3 в Len[inches]: > > >> http://tinyurl.com/yda8nk5x > > >> For 7.5 ft coax, 7.5*12*2/3 = 60 ns. > > >> This is in the ballpark of what he is seeing. > > > That formula's wrong--propagation velocity is about 2nS/ft on FR-4, > > not 2/3rds of a nS per inch! > > > 'c' in vacuum is 3x10^8 m/s, and about 2/3rds that in coax, right? > > > A two-way trip through 7.5' of coax is roughly 5 meters, so > > 5m / (2/3 * 3x10^8 m/s) = 25nS. > > > George's ringing is about 62-ish nS, too slow for the cable to be > > resonant. > > > Cheers, > > James Arthur > > Did you read the article? Eric Bogatin is a well-respected SI expert. He > gives the explanation of ringing coax in the article. I copy it here: > > ------------------------------------------------------------------------- > The root cause of the ringing is reflections from the mismatch at the RX > between the impedance of the line and the high input impedance of the RX, > and the mismatch when the reflected signal makes its way back to the low > output impedance of the TX. > > The reflection coefficient is close to 1 at the RX, but is a negative > value at the TX. Let’s trace the signal path on its journey. When it first > reaches the RX, it reflects. The initial voltage at the RX is high. The > reflected signal makes its way back to the TX, where it sees a high to low > impedance and reflects. But the reflected voltage is negative. > > This negative reflected signal makes its way back to the RX, where it > again reflects, but since it is negative, pulls the signal at the RX down. > When it reflects from the RX, it is still negative, makes its way back to > the TX, changes sign when it reflects, and travels back to the RX as a > positive signal. > > From the initial positive signal at the RX, the signal takes one TD (time > delay) to make it back to the TX, another TD to reflect back to the RX > (that’s 2 × TD from a peak to a valley at the RX), another TD to get to > the TX again, and another TD for the reflected signal to make it to the RX > as a positive signal. > > This is a rather surprising result: four time delays between successive > peaks due to reflections in an unterminated line. > ------------------------------------------------------------------------- > > The difference between Eric's analysis and George's photos is I believe > that George is measuring the ringing at the transmitting end, not at the > unterminated receiving end.No I was looking at the unterminated receiving end... coax plugged into 'scope.> > However, the fact there is no ground plane and the photodode has such high > capacitance could turn the multiple reflections into damped ringing, as > shown in the pictures.Ground plane, smound plane. It's only ~10 MHz, ground comes in on the power supply line, connects to the negative power bypass caps, non-inverting input, goes under opamp to the positve supply caps and coax ground. All told may 1" of trace. (It use to go to the photodiode(s) too, but now reverse biased.) I don't think a ground plane is going to change much. Unless it helps shield the stray FB capacitance. George H.> > Adding a termination resistor to the coax stops the reflections. This > could stop the ringing by eliminating the reflections and also by loading > the op amp with a resistive load.
TIA-PD compensation, by eye
Started by ●October 19, 2017
Reply by ●October 24, 20172017-10-24
Reply by ●October 24, 20172017-10-24
George Herold <gherold@teachspin.com> wrote:>> The difference between Eric's analysis and George's photos is I believe >> that George is measuring the ringing at the transmitting end, not at >> the unterminated receiving end.> No I was looking at the unterminated receiving end... coax plugged into > 'sc ope.Turns out it doesn't matter. The ringing waveform is the same at both ends. The only difference is the amplitude is much smaller at the transmitting end due to the low impedance of the source. Here's the LTSpice files: https://silvercell.000webhostapp.com/sed/4B585018.ZIP> Ground plane, smound plane. It's only ~10 MHz, ground comes in on the > power supply line, connects to the negative power bypass caps, > non-inverting > input, goes under opamp to the positve supply caps and coax ground. > All > told may 1" of trace. (It use to go to the photodiode(s) too, but now > reverse biased.) I don't think a ground plane is going to change much. > Unless it helps shield the stray FB capacitance.> George H.A ground plane gives much shorter bypassing on the VCC and VEE connections. This improves filtering at high frequencies. When you are delivering power to the load, the current flows from the power supply, through the load, then to ground. Any impedance in the power supply connections can produce a signal on the power pins. This can couple into the op amp due to common mode rejection. This is usually much poorer at high frequencies. Ground is the reference for the input to the op amp. Any impedance in the ground connection can produce a voltage at the input to the op amp. The feedback from output to input can produce unwanted resonances, ringing, dips or peaks in the frequency response, and so on. A wideband op amp is much more vulnerable. The datasheet gives specific instructions on bypassing, grounding, layout and other useful information. Note they recommend a sold ground plane for this chip. For bypassing, they state: Minimize the distance (< 0.25�) from the power supply pins to high frequency 0.1-�F and 100-pF de-coupling capacitors. At the device pins, the ground and power plane layout should not be in close proximity to the signal I/O pins. Much more information on page 14 of the datasheet. They also require the thermal pad be soldered to copper, with vias going to the ground plane. Your circuit may be 10 MHz, but this is a 200 MHz device. It is very different from the low frequency op amps you have been using.
Reply by ●October 24, 20172017-10-24
On 2017-10-23 18:32, George Herold wrote:> On Monday, October 23, 2017 at 3:32:03 PM UTC-4, Joerg wrote: >> On 2017-10-23 06:45, George Herold wrote:[...]>> >>> Ground planes and such. Joerg I'm not sure, what I'd like is a big ass >>> photodiode at 1 M ohm gain and 1 MHz BW. (typically ~1 uA of current.) >>> 1 MHz is sorta border line RF, and with 1 Meg ohm a pF of capacitance >>> can ruin your whole day. (I've got hacked pcb's where I learned this... >>> ground plane sanded off around PD and TIA... ugly.) >>> >> >> You can leave open a small island around IN- but this sort of >> requirement asks for Phil's medicine. Cascode and such. > Hmm well I keep talking about the stray feed back capacitance, > which at the moment is the big problem. I'm not sure a ground plane will > solve this problem, but I'll give it a try. > (I was thinking more of a driven shield at the minus input, but > except for high frequency that's almost the same as ground.) > > So I took out a blank pcb today and first measured the trace C. > ( ~1" bus wires into production's SRS RCL meter, 1.7 pF, > it's not all that accurate at the pF level. > I built up the circuit with a 1 Meg R lying flat from the > switch wiper hole, f_3dB = 150kHz, ~1pF. I then soldered > bus wire from the wiper and bent the TH resistor into a > hairpin, (see pic above) f= 120 kHz. I've always wondered > what the C of a hair pin 1/4 W TH R was, about 0.3pF > more than lying flat, is my answer. > > I took out the bus wire and added the switch > f~ 70 kHz, ~2.3 pF, the switch added about 1 pF. > Adding in 9 more hair pin R's brought f_3dB down to > 30 kHz, ~5.3 pF, each R adds another ~0.3 pF, on average. > > Tomorrow I'll try some 0805's. >If you want the lowest capacitance use several resistors in series. That or a really long high voltage version are about the only way to drop down the capacitance.>> >>> I bought a couple of THS4361's so maybe I'll build up a pcb w/o the switch >>> and see what that look like. >>> >> >> Yes, and with a solid ground plane. >> >> >>> Oh the numbers I post upstream are not right. >>> I'm a little bothered by the high gain end, cause I think I was >>> getting more (HF) gain there with a slower opamp, so I need to redo this >>> more carefully. >>> >>> Gain R 3dB >>> 330 20M(Hz) >>> 1k 15M >>> 3.3k 9M >>> 10k 6M >>> 33k 3M >>> 100k 400k >>> 330k 90k >>> 1Meg 30k >>> >> >> At 1M it is bound to fall apart. You might also want to look for amps >> with lower input noise so you don't need such high feedbakc resistor values. > Hmm I need ~1M (or 10M) ohm because of low currents. > Not much light, for some apps > that's because of a weak source, for other atomic physics stuff it's > because light saturates the transitions, (more light, less signal) > For some stuff you have to > turn off the room lights, or hang black cloth's here and there. > you can do lenses and stuff,but for beginners, that's complicated. >Ok but there are other options. For example, you can pick an amp with lower noise for the TIA, use a commensurately lower value for the feedback resistor and follow that TIA with a regular amp that makes up the difference. -- Regards, Joerg http://www.analogconsultants.com/
Reply by ●October 24, 20172017-10-24
On Tuesday, October 24, 2017 at 10:31:49 AM UTC-4, Steve Wilson wrote:> George Herold <gherold@teachspin.com> wrote: > > >> The difference between Eric's analysis and George's photos is I believe > >> that George is measuring the ringing at the transmitting end, not at > >> the unterminated receiving end. > > > No I was looking at the unterminated receiving end... coax plugged into > > 'sc ope. > > Turns out it doesn't matter. The ringing waveform is the same at both > ends. The only difference is the amplitude is much smaller at the > transmitting end due to the low impedance of the source. Here's the > LTSpice files: > > https://silvercell.000webhostapp.com/sed/4B585018.ZIP > > > Ground plane, smound plane. It's only ~10 MHz, ground comes in on the > > power supply line, connects to the negative power bypass caps, > > non-inverting > > input, goes under opamp to the positve supply caps and coax ground. > > All > > told may 1" of trace. (It use to go to the photodiode(s) too, but now > > reverse biased.) I don't think a ground plane is going to change much. > > Unless it helps shield the stray FB capacitance. > > > George H. > > A ground plane gives much shorter bypassing on the VCC and VEE > connections. This improves filtering at high frequencies. > > When you are delivering power to the load, the current flows from the > power supply, through the load, then to ground. > > Any impedance in the power supply connections can produce a signal on the > power pins. This can couple into the op amp due to common mode rejection. > This is usually much poorer at high frequencies. > > Ground is the reference for the input to the op amp. Any impedance in the > ground connection can produce a voltage at the input to the op amp. The > feedback from output to input can produce unwanted resonances, ringing, > dips or peaks in the frequency response, and so on. A wideband op amp is > much more vulnerable. > > The datasheet gives specific instructions on bypassing, grounding, layout > and other useful information. Note they recommend a sold ground plane for > this chip. > > For bypassing, they state: > > Minimize the distance (< 0.25”) from the power supply pins to high > frequency 0.1-µF and 100-pF de-coupling capacitors. At the device pins, > the ground and power plane layout should not be in close proximity to the > signal I/O pins. > > Much more information on page 14 of the datasheet. They also require the > thermal pad be soldered to copper, with vias going to the ground plane. > > Your circuit may be 10 MHz, but this is a 200 MHz device. It is very > different from the low frequency op amps you have been using.Right, look I'm not trying to dispute the use of ground planes. That's my goto option even for 'audio' frequencies. In this case though there's one IC, and a 1" ground trace connecting all the grounds.... how much better is it going to be with a plane? Anyway, I'm going to have to concede a (maybe temporary) defeat. I put surface mount R's everywhere and the BW at 1 Meg ohm was still 40 kHz. (A slight improvement, but not much.) I'm not sure exactly where the C is but it's in the switch. (I left the R's populated, removed the switch, and bus wired the 1 Meg ohm into the circuit. BW=100 kHz a little lower than without the other R's in place (120kHz.) but not much. Oh as a last hack I put 50 ohms in series with the output.. source termination. That cured the cable issue with 330 ohms of FB R. Step response. https://www.dropbox.com/s/coyzt7bkbi7yf5t/PD-330.BMP?dl=0 George H.
Reply by ●October 24, 20172017-10-24
On Tuesday, October 24, 2017 at 10:48:34 AM UTC-4, Joerg wrote:> On 2017-10-23 18:32, George Herold wrote: > > On Monday, October 23, 2017 at 3:32:03 PM UTC-4, Joerg wrote: > >> On 2017-10-23 06:45, George Herold wrote: > > > [...] > > >> > >>> Ground planes and such. Joerg I'm not sure, what I'd like is a big ass > >>> photodiode at 1 M ohm gain and 1 MHz BW. (typically ~1 uA of current.) > >>> 1 MHz is sorta border line RF, and with 1 Meg ohm a pF of capacitance > >>> can ruin your whole day. (I've got hacked pcb's where I learned this... > >>> ground plane sanded off around PD and TIA... ugly.) > >>> > >> > >> You can leave open a small island around IN- but this sort of > >> requirement asks for Phil's medicine. Cascode and such. > > Hmm well I keep talking about the stray feed back capacitance, > > which at the moment is the big problem. I'm not sure a ground plane will > > solve this problem, but I'll give it a try. > > (I was thinking more of a driven shield at the minus input, but > > except for high frequency that's almost the same as ground.) > > > > So I took out a blank pcb today and first measured the trace C. > > ( ~1" bus wires into production's SRS RCL meter, 1.7 pF, > > it's not all that accurate at the pF level. > > I built up the circuit with a 1 Meg R lying flat from the > > switch wiper hole, f_3dB = 150kHz, ~1pF. I then soldered > > bus wire from the wiper and bent the TH resistor into a > > hairpin, (see pic above) f= 120 kHz. I've always wondered > > what the C of a hair pin 1/4 W TH R was, about 0.3pF > > more than lying flat, is my answer. > > > > I took out the bus wire and added the switch > > f~ 70 kHz, ~2.3 pF, the switch added about 1 pF. > > Adding in 9 more hair pin R's brought f_3dB down to > > 30 kHz, ~5.3 pF, each R adds another ~0.3 pF, on average. > > > > Tomorrow I'll try some 0805's. > > > > If you want the lowest capacitance use several resistors in series. That > or a really long high voltage version are about the only way to drop > down the capacitance. > > >> > >>> I bought a couple of THS4361's so maybe I'll build up a pcb w/o the switch > >>> and see what that look like. > >>> > >> > >> Yes, and with a solid ground plane. > >> > >> > >>> Oh the numbers I post upstream are not right. > >>> I'm a little bothered by the high gain end, cause I think I was > >>> getting more (HF) gain there with a slower opamp, so I need to redo this > >>> more carefully. > >>> > >>> Gain R 3dB > >>> 330 20M(Hz) > >>> 1k 15M > >>> 3.3k 9M > >>> 10k 6M > >>> 33k 3M > >>> 100k 400k > >>> 330k 90k > >>> 1Meg 30k > >>> > >> > >> At 1M it is bound to fall apart. You might also want to look for amps > >> with lower input noise so you don't need such high feedbakc resistor values. > > Hmm I need ~1M (or 10M) ohm because of low currents. > > Not much light, for some apps > > that's because of a weak source, for other atomic physics stuff it's > > because light saturates the transitions, (more light, less signal) > > For some stuff you have to > > turn off the room lights, or hang black cloth's here and there. > > you can do lenses and stuff,but for beginners, that's complicated. > > > > Ok but there are other options. For example, you can pick an amp with > lower noise for the TIA, use a commensurately lower value for the > feedback resistor and follow that TIA with a regular amp that makes up > the difference. > > -- > Regards, Joerg > > http://www.analogconsultants.com/Thanks Joerg, It looks like most of the capacitance is in the switch. Why doesn't Phil H. build PD amps with switched gain stages? I've got my answer. :^) George H. (PCB's from advanced circuits are here, so on to some 'real' work.)
Reply by ●October 24, 20172017-10-24
On 10/24/2017 12:59 PM, George Herold wrote:> On Tuesday, October 24, 2017 at 10:48:34 AM UTC-4, Joerg wrote: >> On 2017-10-23 18:32, George Herold wrote: >>> On Monday, October 23, 2017 at 3:32:03 PM UTC-4, Joerg wrote: >>>> On 2017-10-23 06:45, George Herold wrote: >> >> >> [...] >> >>>> >>>>> Ground planes and such. Joerg I'm not sure, what I'd like is a big ass >>>>> photodiode at 1 M ohm gain and 1 MHz BW. (typically ~1 uA of current.) >>>>> 1 MHz is sorta border line RF, and with 1 Meg ohm a pF of capacitance >>>>> can ruin your whole day. (I've got hacked pcb's where I learned this... >>>>> ground plane sanded off around PD and TIA... ugly.) >>>>> >>>> >>>> You can leave open a small island around IN- but this sort of >>>> requirement asks for Phil's medicine. Cascode and such. >>> Hmm well I keep talking about the stray feed back capacitance, >>> which at the moment is the big problem. I'm not sure a ground plane will >>> solve this problem, but I'll give it a try. >>> (I was thinking more of a driven shield at the minus input, but >>> except for high frequency that's almost the same as ground.) >>> >>> So I took out a blank pcb today and first measured the trace C. >>> ( ~1" bus wires into production's SRS RCL meter, 1.7 pF, >>> it's not all that accurate at the pF level. >>> I built up the circuit with a 1 Meg R lying flat from the >>> switch wiper hole, f_3dB = 150kHz, ~1pF. I then soldered >>> bus wire from the wiper and bent the TH resistor into a >>> hairpin, (see pic above) f= 120 kHz. I've always wondered >>> what the C of a hair pin 1/4 W TH R was, about 0.3pF >>> more than lying flat, is my answer. >>> >>> I took out the bus wire and added the switch >>> f~ 70 kHz, ~2.3 pF, the switch added about 1 pF. >>> Adding in 9 more hair pin R's brought f_3dB down to >>> 30 kHz, ~5.3 pF, each R adds another ~0.3 pF, on average. >>> >>> Tomorrow I'll try some 0805's. >>> >> >> If you want the lowest capacitance use several resistors in series. That >> or a really long high voltage version are about the only way to drop >> down the capacitance. >> >>>> >>>>> I bought a couple of THS4361's so maybe I'll build up a pcb w/o the switch >>>>> and see what that look like. >>>>> >>>> >>>> Yes, and with a solid ground plane. >>>> >>>> >>>>> Oh the numbers I post upstream are not right. >>>>> I'm a little bothered by the high gain end, cause I think I was >>>>> getting more (HF) gain there with a slower opamp, so I need to redo this >>>>> more carefully. >>>>> >>>>> Gain R 3dB >>>>> 330 20M(Hz) >>>>> 1k 15M >>>>> 3.3k 9M >>>>> 10k 6M >>>>> 33k 3M >>>>> 100k 400k >>>>> 330k 90k >>>>> 1Meg 30k >>>>> >>>> >>>> At 1M it is bound to fall apart. You might also want to look for amps >>>> with lower input noise so you don't need such high feedbakc resistor values. >>> Hmm I need ~1M (or 10M) ohm because of low currents. >>> Not much light, for some apps >>> that's because of a weak source, for other atomic physics stuff it's >>> because light saturates the transitions, (more light, less signal) >>> For some stuff you have to >>> turn off the room lights, or hang black cloth's here and there. >>> you can do lenses and stuff,but for beginners, that's complicated. >>> >> >> Ok but there are other options. For example, you can pick an amp with >> lower noise for the TIA, use a commensurately lower value for the >> feedback resistor and follow that TIA with a regular amp that makes up >> the difference. >> >> -- >> Regards, Joerg >> >> http://www.analogconsultants.com/ > > Thanks Joerg, It looks like most of the capacitance is in the switch. > Why doesn't Phil H. build PD amps with switched gain stages? > I've got my answer. :^)Yup. As JL put it, "there's no good place for the gain switch in a front end." I did a 1G/50G TIA for an output called Qcept (unfortunately now defunct). It used a non-latching relay to drive the coil of a latching one via capacitors. The coil of the latching relay was bootstrapped to get rid of the ~0.4 pF coil-to-contact capacitance. Interestingly, the 1G resistor had to be kept shorted when not in use, because otherwise its noise current coupled through the capacitance of the open relay contacts and destroyed the measurement on the 50G range. This TIA was for a capacitive surface-voltage probe for detecting sub-monolayer organic contamination on Si wafers. With a PD, one thing you can do is to put TIAs on both ends and switch out the one you aren't using. A relay contact to AC ground doesn't usually make too much of a mess. Cheers Phil Hobbs> > George H. > (PCB's from advanced circuits are here, so on to some 'real' work.) >-- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 hobbs at electrooptical dot net http://electrooptical.net
Reply by ●October 24, 20172017-10-24
On 2017-10-24 09:55, George Herold wrote:> On Tuesday, October 24, 2017 at 10:31:49 AM UTC-4, Steve Wilson wrote: >> George Herold <gherold@teachspin.com> wrote: >> >>>> The difference between Eric's analysis and George's photos is I believe >>>> that George is measuring the ringing at the transmitting end, not at >>>> the unterminated receiving end. >> >>> No I was looking at the unterminated receiving end... coax plugged into >>> 'sc ope. >> >> Turns out it doesn't matter. The ringing waveform is the same at both >> ends. The only difference is the amplitude is much smaller at the >> transmitting end due to the low impedance of the source. Here's the >> LTSpice files: >> >> https://silvercell.000webhostapp.com/sed/4B585018.ZIP >> >>> Ground plane, smound plane. It's only ~10 MHz, ground comes in on the >>> power supply line, connects to the negative power bypass caps, >>> non-inverting >>> input, goes under opamp to the positve supply caps and coax ground. >>> All >>> told may 1" of trace. (It use to go to the photodiode(s) too, but now >>> reverse biased.) I don't think a ground plane is going to change much. >>> Unless it helps shield the stray FB capacitance. >> >>> George H. >> >> A ground plane gives much shorter bypassing on the VCC and VEE >> connections. This improves filtering at high frequencies. >> >> When you are delivering power to the load, the current flows from the >> power supply, through the load, then to ground. >> >> Any impedance in the power supply connections can produce a signal on the >> power pins. This can couple into the op amp due to common mode rejection. >> This is usually much poorer at high frequencies. >> >> Ground is the reference for the input to the op amp. Any impedance in the >> ground connection can produce a voltage at the input to the op amp. The >> feedback from output to input can produce unwanted resonances, ringing, >> dips or peaks in the frequency response, and so on. A wideband op amp is >> much more vulnerable. >> >> The datasheet gives specific instructions on bypassing, grounding, layout >> and other useful information. Note they recommend a sold ground plane for >> this chip. >> >> For bypassing, they state: >> >> Minimize the distance (< 0.25”) from the power supply pins to high >> frequency 0.1-µF and 100-pF de-coupling capacitors. At the device pins, >> the ground and power plane layout should not be in close proximity to the >> signal I/O pins. >> >> Much more information on page 14 of the datasheet. They also require the >> thermal pad be soldered to copper, with vias going to the ground plane. >> >> Your circuit may be 10 MHz, but this is a 200 MHz device. It is very >> different from the low frequency op amps you have been using. > > Right, look I'm not trying to dispute the use of ground planes. That's > my goto option even for 'audio' frequencies.Actually it's the opposite. Audio is the only area where star-ground (sometimes) works. However, only in tube amps or if everything is CMOS.> ... In this case though there's > one IC, and a 1" ground trace connecting all the grounds.... how much > better is it going to be with a plane? >In the real world with RF-pollution by cell phones, SMPS and all, a lot better.> Anyway, I'm going to have to concede a (maybe temporary) defeat. I put > surface mount R's everywhere and the BW at 1 Meg ohm was still 40 kHz. > (A slight improvement, but not much.) I'm not sure exactly where the C > is but it's in the switch. (I left the R's populated, removed the switch, > and bus wired the 1 Meg ohm into the circuit. BW=100 kHz a little lower > than without the other R's in place (120kHz.) but not much. >If you put 3-5 resistors in series shaped as an arch it'll get even better. But I doubt you'd milk 1MHz out of it. Better to reduce the resistance, pick a lower noise amp and then bump up the gain via a 2nd stage.> Oh as a last hack I put 50 ohms in series with the output.. source > termination. That cured the cable issue with 330 ohms of FB R. > Step response. > > https://www.dropbox.com/s/coyzt7bkbi7yf5t/PD-330.BMP?dl=0 >That still shows serious ringing, probably only borderline stable. I would not release that into the wild yonder as a product. -- Regards, Joerg http://www.analogconsultants.com/
Reply by ●October 24, 20172017-10-24
On Tuesday, October 24, 2017 at 2:58:41 PM UTC-4, Phil Hobbs wrote:> On 10/24/2017 12:59 PM, George Herold wrote: > > On Tuesday, October 24, 2017 at 10:48:34 AM UTC-4, Joerg wrote: > >> On 2017-10-23 18:32, George Herold wrote: > >>> On Monday, October 23, 2017 at 3:32:03 PM UTC-4, Joerg wrote: > >>>> On 2017-10-23 06:45, George Herold wrote: > >> > >> > >> [...] > >> > >>>> > >>>>> Ground planes and such. Joerg I'm not sure, what I'd like is a big ass > >>>>> photodiode at 1 M ohm gain and 1 MHz BW. (typically ~1 uA of current.) > >>>>> 1 MHz is sorta border line RF, and with 1 Meg ohm a pF of capacitance > >>>>> can ruin your whole day. (I've got hacked pcb's where I learned this... > >>>>> ground plane sanded off around PD and TIA... ugly.) > >>>>> > >>>> > >>>> You can leave open a small island around IN- but this sort of > >>>> requirement asks for Phil's medicine. Cascode and such. > >>> Hmm well I keep talking about the stray feed back capacitance, > >>> which at the moment is the big problem. I'm not sure a ground plane will > >>> solve this problem, but I'll give it a try. > >>> (I was thinking more of a driven shield at the minus input, but > >>> except for high frequency that's almost the same as ground.) > >>> > >>> So I took out a blank pcb today and first measured the trace C. > >>> ( ~1" bus wires into production's SRS RCL meter, 1.7 pF, > >>> it's not all that accurate at the pF level. > >>> I built up the circuit with a 1 Meg R lying flat from the > >>> switch wiper hole, f_3dB = 150kHz, ~1pF. I then soldered > >>> bus wire from the wiper and bent the TH resistor into a > >>> hairpin, (see pic above) f= 120 kHz. I've always wondered > >>> what the C of a hair pin 1/4 W TH R was, about 0.3pF > >>> more than lying flat, is my answer. > >>> > >>> I took out the bus wire and added the switch > >>> f~ 70 kHz, ~2.3 pF, the switch added about 1 pF. > >>> Adding in 9 more hair pin R's brought f_3dB down to > >>> 30 kHz, ~5.3 pF, each R adds another ~0.3 pF, on average. > >>> > >>> Tomorrow I'll try some 0805's. > >>> > >> > >> If you want the lowest capacitance use several resistors in series. That > >> or a really long high voltage version are about the only way to drop > >> down the capacitance. > >> > >>>> > >>>>> I bought a couple of THS4361's so maybe I'll build up a pcb w/o the switch > >>>>> and see what that look like. > >>>>> > >>>> > >>>> Yes, and with a solid ground plane. > >>>> > >>>> > >>>>> Oh the numbers I post upstream are not right. > >>>>> I'm a little bothered by the high gain end, cause I think I was > >>>>> getting more (HF) gain there with a slower opamp, so I need to redo this > >>>>> more carefully. > >>>>> > >>>>> Gain R 3dB > >>>>> 330 20M(Hz) > >>>>> 1k 15M > >>>>> 3.3k 9M > >>>>> 10k 6M > >>>>> 33k 3M > >>>>> 100k 400k > >>>>> 330k 90k > >>>>> 1Meg 30k > >>>>> > >>>> > >>>> At 1M it is bound to fall apart. You might also want to look for amps > >>>> with lower input noise so you don't need such high feedbakc resistor values. > >>> Hmm I need ~1M (or 10M) ohm because of low currents. > >>> Not much light, for some apps > >>> that's because of a weak source, for other atomic physics stuff it's > >>> because light saturates the transitions, (more light, less signal) > >>> For some stuff you have to > >>> turn off the room lights, or hang black cloth's here and there. > >>> you can do lenses and stuff,but for beginners, that's complicated. > >>> > >> > >> Ok but there are other options. For example, you can pick an amp with > >> lower noise for the TIA, use a commensurately lower value for the > >> feedback resistor and follow that TIA with a regular amp that makes up > >> the difference. > >> > >> -- > >> Regards, Joerg > >> > >> http://www.analogconsultants.com/ > > > > Thanks Joerg, It looks like most of the capacitance is in the switch. > > Why doesn't Phil H. build PD amps with switched gain stages? > > I've got my answer. :^) > > Yup. As JL put it, "there's no good place for the gain switch in a > front end." I did a 1G/50G TIA for an output called Qcept > (unfortunately now defunct). It used a non-latching relay to drive the > coil of a latching one via capacitors. The coil of the latching relay > was bootstrapped to get rid of the ~0.4 pF coil-to-contact capacitance. > > Interestingly, the 1G resistor had to be kept shorted when not in use, > because otherwise its noise current coupled through the capacitance of > the open relay contacts and destroyed the measurement on the 50G range. > > This TIA was for a capacitive surface-voltage probe for detecting > sub-monolayer organic contamination on Si wafers. > > With a PD, one thing you can do is to put TIAs on both ends and switch > out the one you aren't using. A relay contact to AC ground doesn't > usually make too much of a mess.Hmm I think you mentioned the two TIA trick before. I was thinking that I could make a series string of gain resistors rather than the parallel connection I have now. I drew that up and if I've got the stray C in the right place, then I think it will be a little better. (factor of two or so.) 'cause now at the low R end the stray C is shunting a smaller resistor, doing less harm. George H.>> Cheers > > Phil Hobbs > > > > > > George H. > > (PCB's from advanced circuits are here, so on to some 'real' work.) > > > > > -- > Dr Philip C D Hobbs > Principal Consultant > ElectroOptical Innovations LLC > Optics, Electro-optics, Photonics, Analog Electronics > > 160 North State Road #203 > Briarcliff Manor NY 10510 > > hobbs at electrooptical dot net > http://electrooptical.net
Reply by ●October 24, 20172017-10-24
On Tuesday, October 24, 2017 at 4:03:41 PM UTC-4, Joerg wrote:> On 2017-10-24 09:55, George Herold wrote: > > On Tuesday, October 24, 2017 at 10:31:49 AM UTC-4, Steve Wilson wrote: > >> George Herold <gherold@teachspin.com> wrote: > >> > >>>> The difference between Eric's analysis and George's photos is I believe > >>>> that George is measuring the ringing at the transmitting end, not at > >>>> the unterminated receiving end. > >> > >>> No I was looking at the unterminated receiving end... coax plugged into > >>> 'sc ope. > >> > >> Turns out it doesn't matter. The ringing waveform is the same at both > >> ends. The only difference is the amplitude is much smaller at the > >> transmitting end due to the low impedance of the source. Here's the > >> LTSpice files: > >> > >> https://silvercell.000webhostapp.com/sed/4B585018.ZIP > >> > >>> Ground plane, smound plane. It's only ~10 MHz, ground comes in on the > >>> power supply line, connects to the negative power bypass caps, > >>> non-inverting > >>> input, goes under opamp to the positve supply caps and coax ground. > >>> All > >>> told may 1" of trace. (It use to go to the photodiode(s) too, but now > >>> reverse biased.) I don't think a ground plane is going to change much. > >>> Unless it helps shield the stray FB capacitance. > >> > >>> George H. > >> > >> A ground plane gives much shorter bypassing on the VCC and VEE > >> connections. This improves filtering at high frequencies. > >> > >> When you are delivering power to the load, the current flows from the > >> power supply, through the load, then to ground. > >> > >> Any impedance in the power supply connections can produce a signal on the > >> power pins. This can couple into the op amp due to common mode rejection. > >> This is usually much poorer at high frequencies. > >> > >> Ground is the reference for the input to the op amp. Any impedance in the > >> ground connection can produce a voltage at the input to the op amp. The > >> feedback from output to input can produce unwanted resonances, ringing, > >> dips or peaks in the frequency response, and so on. A wideband op amp is > >> much more vulnerable. > >> > >> The datasheet gives specific instructions on bypassing, grounding, layout > >> and other useful information. Note they recommend a sold ground plane for > >> this chip. > >> > >> For bypassing, they state: > >> > >> Minimize the distance (< 0.25”) from the power supply pins to high > >> frequency 0.1-µF and 100-pF de-coupling capacitors. At the device pins, > >> the ground and power plane layout should not be in close proximity to the > >> signal I/O pins. > >> > >> Much more information on page 14 of the datasheet. They also require the > >> thermal pad be soldered to copper, with vias going to the ground plane. > >> > >> Your circuit may be 10 MHz, but this is a 200 MHz device. It is very > >> different from the low frequency op amps you have been using. > > > > Right, look I'm not trying to dispute the use of ground planes. That's > > my goto option even for 'audio' frequencies. > > > Actually it's the opposite. Audio is the only area where star-ground > (sometimes) works. However, only in tube amps or if everything is CMOS. > > > > ... In this case though there's > > one IC, and a 1" ground trace connecting all the grounds.... how much > > better is it going to be with a plane? > > > > In the real world with RF-pollution by cell phones, SMPS and all, a lot > better. > > > > Anyway, I'm going to have to concede a (maybe temporary) defeat. I put > > surface mount R's everywhere and the BW at 1 Meg ohm was still 40 kHz. > > (A slight improvement, but not much.) I'm not sure exactly where the C > > is but it's in the switch. (I left the R's populated, removed the switch, > > and bus wired the 1 Meg ohm into the circuit. BW=100 kHz a little lower > > than without the other R's in place (120kHz.) but not much. > > > > If you put 3-5 resistors in series shaped as an arch it'll get even > better. But I doubt you'd milk 1MHz out of it. Better to reduce the > resistance, pick a lower noise amp and then bump up the gain via a 2nd > stage. > > > > Oh as a last hack I put 50 ohms in series with the output.. source > > termination. That cured the cable issue with 330 ohms of FB R. > > Step response. > > > > https://www.dropbox.com/s/coyzt7bkbi7yf5t/PD-330.BMP?dl=0 > > > > That still shows serious ringing, probably only borderline stable. I > would not release that into the wild yonder as a product.Oh the slow yellow curve (Chan 1) is the PD. The blue ringy thing is the drive current into the LED. Not for release to public, well I might hack a 'one of' to send to a valued customer. Nor for sale. George H.> > -- > Regards, Joerg > > http://www.analogconsultants.com/
Reply by ●October 24, 20172017-10-24
On 2017-10-24 13:25, George Herold wrote:> On Tuesday, October 24, 2017 at 4:03:41 PM UTC-4, Joerg wrote: >> On 2017-10-24 09:55, George Herold wrote:[...]>>> Oh as a last hack I put 50 ohms in series with the output.. source >>> termination. That cured the cable issue with 330 ohms of FB R. >>> Step response. >>> >>> https://www.dropbox.com/s/coyzt7bkbi7yf5t/PD-330.BMP?dl=0 >>> >> >> That still shows serious ringing, probably only borderline stable. I >> would not release that into the wild yonder as a product. > > Oh the slow yellow curve (Chan 1) is the PD. The blue ringy > thing is the drive current into the LED. >The yellow response looks fine. The noise is probably there because it's one of those Tek lunch-box sized scopes (the reason I bought a GW-Instek instead). The BW is quite low though but I you work on reducing feedback capacitances that should improve.> Not for release to public, well I might hack a 'one of' to > send to a valued customer. Nor for sale. >That's good. Most of my clients have test customers who will gladly try out stuff. -- Regards, Joerg http://www.analogconsultants.com/
