The datasheet only goes down to 10kHz and I need the response down to sub Hz

The datasheet only goes down to 10kHz and I need the response down to sub Hz

Dear Phil,

I am somewhat surprised that you do not have the low frequency response data as I would have expected that to be central to the original development of the ADA4530-1. I am developing very high gain transimpedance amplifiers (100Tohm) for application in mass spectrometers, in particular for measurement of geochronological times i.e. billions of years (see e.g. argon-argon dating in Wikipedia). The response time of these amplifiers has to be much faster than you would expect from such high value feedback resistors (the ions disappear quite quickly in the spectrometer vacuum) so a number of special techniques have been developed to improve the response time to a few seconds. Part of the theoretical analysis of the circuits requires the knowledge of the zero-frequency gain and of the open-loop corner frequency, and hence the request I have made. The ADA4530 looks particularly attractive as the input amplifier for its very low bias current, which is the limiting factor in the ultimate signai-to-noise ratio, and for the built-in guard driving amplifier. The bandwidth of the system is of the order of 10mHz to 1Hz so well within the presumed open-loop bandwidth of the 4530. To obtain even better precision the 4530 is followed within the loop by a ~X10 non-inverting stage (OP37) so achieving dynamic stability does present some difficulty in achieving a suitable open-loop frequency response. A major reason for trying to achieve such high gain transimpedance amplifiers is of course to be able to measure the tiny atto or femtoamp ion currents to determine the isotopic ratios of say Argon (you do not get much Argon in a piece of meteorite or moon rock). The general argument is that increasing the feedback resistor by say 10X increases the signal gain by the same ratio but the noise, which is primarily the Johnson noise of the feedback resistor, only increases by root 10 so the S/N ratio increases by root 10. The ultimate limit is reached when the shot noise of the amplifier bias current becomes greater than the Johnson noise. To obtain the best performance the amplifiers are housed in an evacuated and temperature controlled housing to avoid surface contamination of the feedback resistor and their significant temperature coefficients. These high value resistors leave much to be desired.

I hope this gives you enough information as to our application and that you will be able to provide us with the requested information. If you would like any further information please let me know.

Yours sincerely, Dr Scott Hamilton.

TIA Systems

Hello Scott,

Thank you very much for the very detailed explanation of your application and interest to use our products. We're so glad that our amplifiers are being used in this very exciting field of applications. In order to help you better, someone from the original development team of the ADA4530-1 will handle your question and get back to you as soon as possible.

Thanks again and best regards,

Phil

Dr. Hamilton,

Thank you for your detailed explanation of your work. It is very exciting to see our products considered for such high end measurement systems.Let me start by offering some context about how we present gain on our datasheets.

The first topic is the reason for the 10 kHz starting frequency in the open-loop gain vs. frequency graph (Figure 55).

This is done to try to maximize the information conveyed by the graph. The majority of users are concerned with stabilizing their feedback loops. The amplifier phase shift and gain near unity provide useful details that allow users to predict the phase margin and gain margin of their loops using their feedback functions of interest. If we were to ‘zoom out’ the graph by providing more low frequency data, these details would become more obscured.

Having said that, there is no reason why we could not have two graphs; the complete response and the detailed response. I do not think that we have ever had any requests for this before. To date, I am not aware of anyone else who has attempted to construct a compensator for the low frequency amplifier pole.

One potential problem with this is that the DC gain and dominate pole frequency are not well controlled. The gain-bandwidth product is controlled within 20% or so because of the good tolerance of the FET transconductance and compensation capacitance. The DC gain and dominate pole are affected by the output resistance of the cascaded FETs in the gain stages. There will be significant part-to-part variation in this parameter as well as a large temperature coefficient.

Now let’s discuss actually measuring the dominate pole. From large signal voltage gain measurements, the gain of the ADA4530 is typically 150-160 dB (Note that the 125 dB minimum in the data sheet is a limitation of production test, not representative of the part). The gain-bandwidth product of 2 MHz suggests a dominate pole frequency 20 mHz.

Now consider the noise requirements. Assume that output swing is 9 Vpp (on a +/- 5 V power supply). The input error signal will be 90 nVpp (32 nVrms). The 1/f noise spectral density is 2 uV/sqrt(Hz) at 20 mHz (extrapolated from Figure 85). I calculate a measurement bandwidth of 256 uHz to get an integrated noise of 32 nVrms. This back of the envelope calculation also ignores the 1/f nature of the noise which makes it very optimistic.

This measurement bandwidth will require slightly over 1 hour to measure a single point (at a SNR around 1). This presents another challenge: the temperature must be extremely well controlled over this time period to coupling in offset voltage drift (TCVos) errors into the measurement. Even with the excellent TCVos (130 nV/C typ) of the amplifier, it would be challenging to suppress this error. For example, reducing this error to 13 nV (1/3 the signal) requires temperature stability of 10 mK. This level of stability is far beyond our ambient temperature stability or even the abilities of our ovens. I think that we would have to make this measurement in an oil bath.

In summary, this is a very challenging measurement to make which requires many days to complete all of the averages and frequency points. My concern is that at the end of this the unit-to-unit variation is large enough to render the measurement useless.

Perhaps I have misunderstood something. If I have, please correct me.

Regards,

Mark ReisigerDear Mark,

Many thanks for your extensive reply and explanation of the difficulty of open-loop measurement on the ADA4530. I had simply assumed that such information would have been part of the development of the device. I do not wish to get you involved in such a lengthy measurement and will try to use the approximate values you have quoted i.e. zero frequency gain ~150dB and corner frequency ~20mHz. The latter is right in the region of my 100Tohm transimpedance amplifier response. It has been fairly easy to get 'fast' risetime (compared with what is generally expected from the value of the feedback resistor and parasitic capacity) but the most difficult part is to get output settling to the final level within the noise level within a few seconds. Transients during settling are difficult to control. Once the samples are introduced into the spectrometer they decay rapidly (vacuum pumping) and more concerning is the possible differential decay between different isotopes of the same element so the ratios have to be measured quickly.

Thank you again for your assistance.

Kind regards, Scott Hamilton.

Dr. Hamilton,

With regard to the ‘fast rise-time but slow settling’ behavior, I have a few ideas.The first observation is that this behavior may be caused by your compensator. Introducing a zero to cancel out the frequency limiting pole will always introduce a pole-zero doublet in the system. Pole-zero doublets will rise extremely fast to a value equal to the ratio of the pole and zero. After this, the final settling occurs with the time constant of the pole which is much slower. Unfortunately, it is never possible to match the pole and zero exactly. As the system accuracy requirements increase, this implies that the pole-zero cancelation must increase by the same amount. This can be very difficult to achieve in the presence of temperature fluctuations, etc.

One other possible cause of slow (long tail) settling behavior in very high impedance systems is dielectric relaxation. I have observed that even the best insulators can contribute 10’s of femtoamperes of dielectric relaxation current with time constants of many seconds. Fortunately, transimpedance amplifiers suppress these effects because the high impedance nodes are held to virtual ground potentials. This favorable situation breaks down if the TIA is over-ranged. If this happens, the summing junction voltage changes and the recovery time can be extremely long because the dielectric relaxation must settle out.

Have you ever considered using capacitive integration techniques? This approach is much more responsive to high frequency events than the TIA. In addition, the system noise is lower without the thermal noise of the feedback resistor. The downside is the added complication of needing to reset the integration. Fortunately, the ADA4530 has such low bias current that the reset interval can be infrequent. I have used the ADA4530 in this manner with a 1 pF integration capacitor that is reset with a guarded reed relay. This system can operate for hours between resets (of course any signal current that you add will speed this up).

Regards,

Mark Reisiger

Dr. Hamilton,

Thank you for your detailed explanation of your work. It is very exciting to see our products considered for such high end measurement systems.

Let me start by offering some context about how we present gain on our datasheets.

The first topic is the reason for the 10 kHz starting frequency in the open-loop gain vs. frequency graph (Figure 55).

This is done to try to maximize the information conveyed by the graph. The majority of users are concerned with stabilizing their feedback loops. The amplifier phase shift and gain near unity provide useful details that allow users to predict the phase margin and gain margin of their loops using their feedback functions of interest. If we were to ‘zoom out’ the graph by providing more low frequency data, these details would become more obscured.

Having said that, there is no reason why we could not have two graphs; the complete response and the detailed response. I do not think that we have ever had any requests for this before. To date, I am not aware of anyone else who has attempted to construct a compensator for the low frequency amplifier pole.

One potential problem with this is that the DC gain and dominate pole frequency are not well controlled. The gain-bandwidth product is controlled within 20% or so because of the good tolerance of the FET transconductance and compensation capacitance. The DC gain and dominate pole are affected by the output resistance of the cascaded FETs in the gain stages. There will be significant part-to-part variation in this parameter as well as a large temperature coefficient.

Now let’s discuss actually measuring the dominate pole. From large signal voltage gain measurements, the gain of the ADA4530 is typically 150-160 dB (Note that the 125 dB minimum in the data sheet is a limitation of production test, not representative of the part). The gain-bandwidth product of 2 MHz suggests a dominate pole frequency 20 mHz.

Now consider the noise requirements. Assume that output swing is 9 Vpp (on a +/- 5 V power supply). The input error signal will be 90 nVpp (32 nVrms). The 1/f noise spectral density is 2 uV/sqrt(Hz) at 20 mHz (extrapolated from Figure 85). I calculate a measurement bandwidth of 256 uHz to get an integrated noise of 32 nVrms. This back of the envelope calculation also ignores the 1/f nature of the noise which makes it very optimistic.

This measurement bandwidth will require slightly over 1 hour to measure a single point (at a SNR around 1). This presents another challenge: the temperature must be extremely well controlled over this time period to coupling in offset voltage drift (TCVos) errors into the measurement. Even with the excellent TCVos (130 nV/C typ) of the amplifier, it would be challenging to suppress this error. For example, reducing this error to 13 nV (1/3 the signal) requires temperature stability of 10 mK. This level of stability is far beyond our ambient temperature stability or even the abilities of our ovens. I think that we would have to make this measurement in an oil bath.

In summary, this is a very challenging measurement to make which requires many days to complete all of the averages and frequency points. My concern is that at the end of this the unit-to-unit variation is large enough to render the measurement useless.

Perhaps I have misunderstood something. If I have, please correct me.

Regards,

Mark Reisiger