Auto Nulling of Operational Amplifiers within FlexAnalog FPAAs
App Note: 302 (636 kB, 1 pdf file): Auto Nulling of Operational Amplifiers
Introduction
The presence of offsets in operational amplifiers is an unfortunate fact of life. These offsets become particularly troublesome in implementations with high gain where the offset of the OpAmp is, approximately speaking, multiplied by the gain of the circuit. In such cases, the resulting output offset can be enough to limit the available signal swing, or can even saturate the OpAmp in extreme cases. This effect can occur in unexpected places as well as gain stages. Some filter configurations, for example, have high gains internally which can detrimentally interact with OpAmp offsets to give poor overall circuit performance.
The chopping schemes in the CAB OpAmp (and chopping amplifier itself) do not operate well with a large intrinsic OpAmp offset. The combination of large gain, significant OpAmp offset and chopping may lead to the production of a large square wave at the OpAmp output. While this is usually at a frequency high enough not to be problematic, it can lead to settling problems in subsequent circuits.
Circuit topologies exist which do a lot to deal with OpAmp offsets. Half cycle structures are often employed to compensate for OpAmp offset, but these are not always possible to deploy in a given situation. The programmable nature of the FlexAnalogTM FPAAs means it is practically impossible to compensate for the effects of OpAmp offset in all possible situations, so FlexAnalogTM FPAAs deal with offsets by greatly reducing them at the source.
FlexAnalogTM FPAAs contain an auto-null scheme to reduce the input offset of the CAB, IO and chopper amplifiers. This document explains what it is and how to use it by either configuration or re-configuration. The document has two main sections – the first is a detailed description of auto-nulling, the second is a practical guide on how to use it.
Detailed Description of Auto-nulling
1.1 Auto-null offset correction
The auto-null scheme is a well-known and widely used technique to reduce OpAmp offsets. Essentially it works by injecting a current into the OpAmp which acts to compensate for the intrinsic offset. The resulting OpAmp has an offset very much reduced from its intrinsic value.
An auto-null scheme has been implemented in the complex I/O and the CAB OpAmps so that a low offset signal path can be made from input to output pin in the FPAA. A chopping amplifier is also provided in the IO which can give significant gain and both auto-null and chopping offset correction; this is for situations where the input signal is small and needs the application of a large gain before signal processing elsewhere.
The FPAA auto-null schemes operate on the differential offsets only. The common mode offsets are left uncorrected because they are not multiplied by user circuit topology. Common mode offsets are rejected by default by the differential nature of the FPAA.
1.2 Details of the auto-null implementation
Activation of the auto-null schemes is possible in some ways, as detailed below. These methods differ only regarding the user’s programming choice; the mechanisms within the chip are the same. Throughout this document “OpAmps” refers generically to the operational amplifiers in the complex IO, the CAB or the chopper amplifier.
1.2.1 Primary configuration
The auto-null implementation is the primary usage mechanism. Auto-null is selectively activated for all OpAmps which are powered up for use in a given application. Other OpAmps are left powered down with their intrinsic offsets still present. After configuration and the subsequent auto-null is complete, the OpAmps come alive, and signal processing begins.
Once an auto-null is completed for an OpAmp, the offset correction values are stored in local RAM. This will retain its values until the device is reset, power is cycled, or auto-nulling is disabled. The RAM is not part of the memory map and is local to each OpAmp.
1.2.2 Reconfiguration
After a primary configuration, it may be required to periodically re-run the auto-null if operating conditions (e.g. temperature) have significantly changed. Changes in operating conditions can cause the OpAmp offset to drift and the compensation previously applied may then no longer be correct. It may also be required to power up OpAmps which were not previously auto-nulled, and then activate the auto-null scheme for them. Both these options are completed via re-configuration.
1.2.3 Suspended animation
It is possible to perform an auto-null on an OpAmp (either via primary or re-configuration) then power the OpAmp down. When such an OpAmp is subsequently powered up it will have retained its low offset values, provided there have been no significant changes in operating conditions and provided that there has been no power cycle, reset or disabling of auto-null.
1.2.4 No auto-null
By default, auto-null is on but can be disabled easily, and may also be de-activated after being deployed. Either of these states will yield an OpAmp which displays its intrinsic offset voltage. Although this eradicates the delay required to auto-null before an OpAmp can be used, it is probably of limited value.
1.3 OpAmp circuit details
Autozero schemes exist in the IO, CAB and chopper amplifiers. These all have the same basic architecture, where a test current is injected into the differential input stage, as per Figure 1 below.
The magnitude and polarity of the current necessary to cancel out the OpAmp offset- voltage is determined by a successive-approximation routine.
1.4 Auto-null User Requirements
The auto-null cycle for any OpAmp requires some external conditions to be met requiring a configuration from the user and an external clock. Necessary conditions are:
- Providing an external clock on either ACLK or DCLK.
- Setting a clock divider to give a suitably slow internal clock.
- Setting a bit(s) to enable the auto calibration.
- Waiting for the specified duration before expecting valid functionality from the OpAmp undergoing auto-null. OpAmps which are not set to run an auto-null cycle will continue signal processing.
The external clock is required to drive the state machine(s) which operate the auto-null schemes, and it must be suitably divided down to around 100Hz internally. A special clock divider and associated dedicated routing are provided for this purpose, so no special connectivity is required from the user. The auto-null clock has a fixed divider by 1000 as a pre-scaler and a subsequent post-scaler to allow the user finer control of the final auto-null clock frequency.
The FlexAnalogTM auto-null scheme uses the OpAmps in open loop mode, where they may respond at a frequency equal to their dominant pole frequency. 100Hz is regarded as a safe value. Higher frequencies may be used with the increased risk that the auto-null result will not be as small as it otherwise would be, due to incomplete OpAmp settling.
The required action from the user can all be accomplished in a single primary configuration, including any other required circuit functions if desired. An example is given in Appendix A.
1.5 End of auto-null indication
The auto-null scheme has an explicit way of signaling externally that it has finished. Pins LCC_B, MEMSETUP or MEMCLK can be programmed to give this function once the device is configured.
1.6 Chopping and auto-null interactions
The use of chopping is a good way to render the offset of an OpAmp to appear very close to null. While chopping is active, the differential output of an OpAmp switches between +offset_voltage and -offset_voltage at the chopping frequency. If large gains are applied, then the output approximately switches between
+Gain*offset_voltage and -Gain*offset_voltage which can potentially be several 100mV or more. The chopped amplifier will continue to operate under these conditions, but the large square wave being driven out causes settling time problems for this and other circuits and is also a reasonably strong noise coupling source.
For these reasons, it is highly desirable to auto-null any chopped amplifier. Thus, the effective magnitude of the offset_voltage quantity is much smaller, and the settling and noise problems are greatly reduced.
2. Practical Guide to Auto-nulling
2.1 Set-up in Anadigm Designer 2 Software
Fortunately, the Anadigm Designer 2 Software makes it very simple to use auto-nulling. All the user has to do is enable it by doing the following:
- Right-click over the circuit
- Select the Chip tab
- Check the box labeled “Use Offset Nulling.”
This is all the user has to do since the Anadigm Designer 2 Software will now create the appropriate configuration data to enable offset nulling and calculate the required nulling clock divider to give the correct frequency for the nulling clock (~100Hz). In fact, the user doesn’t even need to do this because auto-nulling is enabled by default.
There are some points to note, however:
- Only OpAmps that are actually used in the circuit will be auto-nulled.
- Once auto-nulled, OpAmps will stay auto-nulled until reset, power down or auto-nulling is disabled.
- Auto-nulling requires a delay of approximately 60ms immediately after configuration, during which time the OpAmps being auto-nulled cannot be used (see section 2.3).
Regarding point “a.” If a circuit is loaded with a primary configuration in which auto-nulling is enabled. And that circuit does not use every OpAmp. Then if the user wishes to do a re-configuration (i.e., an update, not a primary configuration) to another circuit which enables previously unused OpAmps, then those newly enabled OpAmps will not have auto-nulling enabled. In this situation, the user must restart auto-nulling. This can be done in one of the following ways:
- If the reconfigured circuit has auto-nulling disabled, after the reconfiguration has been sent to the FPAA a second short auto-null reconfiguration is sent that turns auto-nulling back on. This forces auto-nulling to be run again for all of the OpAmps in the new circuit. The short auto-null reconfiguration is shown in Appendix B.
- The alternative is to reset the circuit and send the second circuit as a primary configuration, with auto-nulling enabled of course. This will have the additional delay of a reset as well as a new auto-nulling sequence.
NOTE: the short auto-null reconfiguration mentioned in i. above and described in Appendix B can be sent to the FPAA after a primary configuration or reconfiguration in which auto-nulling was disabled. This will cause auto-nulling to be executed. However, if the preceding configuration had auto-nulling enabled, the short auto-null reconfiguration will not enable auto-nulling.
If the user cannot tolerate any delay when changing to the new circuit, and if offsets are not critical in the new circuit, then the user could disable auto-nulling in the reconfiguration. This will mean that the circuit becomes functional immediately, but the OpAmps will have their (larger) intrinsic offsets.
Regarding point “c.” on the previous page. It is important to note that auto-nulling incurs a delay and that OpAmps will be nonfunctional while being auto-nulled. The user should therefore ensure that there is a delay of 60ms after configuration before the circuit is used. It is possible to enable the FPAA to assert a signal when auto-nulling is complete (see Figure 3). To enable an auto-nulling done signal, double- click on the digital IO cell in Anadigm Designer 2 Software and select Auto-null Done for either Dout1 (pin 39) or Dout2 (pin 42).
Figure 3 shows the digital IO cell with both signals enabled. Note that pin 39 is also called LCC and pin 42 is also called MEMCLK.
2.2 Auto-nulling Delay
Figure 4 shows waveforms for a primary configuration of a circuit containing a single GainInv CAM. The bottom trace shows SCLK. After completion of configuration (SCLK stops pulsing), the auto-null DONE signal goes low, and the analog output goes to VMR, but there is no output signal. After about 60ms the analog output starts, and after that, the auto-null DONE signal is asserted.
It can be seen that the done signal does not get asserted immediately after the OpAmp starts to work. The purpose of the auto-null DONE signal is simply to provide the user with a signal to tell the rest of the system that it is safe to use the analog circuit in the FPAA. The user can either wait 60ms or use the done signal.
The exact delay in completion of auto-nulling and the assertion of the auto-null done signal can be calculated from the master clock frequency and the auto-nulling clock divider. As stated previously, the auto-nulling clock is obtained by dividing the master clock by a fixed pre-scaler of 1000 and a variable secondary divider. The secondary divider is calculated by the Software to give an auto-null clock of as near to 100Hz as possible. The Software puts the secondary divider value into byte 14, bank 0 of the configuration data (see Appendix A). Note that clock dividers only have even values except for 1, so to get the actual divisor value multiply the configuration byte by 2 (a configuration byte of 0 means a divisor of 1). Finally, it is necessary to know that auto-nulling is complete in 6 auto-null clocks and the auto-nulling done signal is asserted in 10 auto-null clocks.
So the formulae for auto-nulling delays are:
- Auto-null complete delay = 6 x 1000 x Div / Fmaster
- Assert done signal delay = 10 x 1000 x Div / Fmaster
- Where Div is the auto-nulling secondary divider and Fmaster is the frequency of the master clock.
NOTE: the clock dividers take effect before the end of the configuration so if the delay is taken from the end of the configuration byte stream these delays will be slightly shorter.
2.3 Distribution of the offset voltages
The intrinsic offset voltages follow a Gaussian distribution, centered around a mean value. Ideally, the mean is zero, but it is often a few mV away from that due to imperfections in the layout of differential circuitry. After auto-nulling, the offset voltage is within the given bounds, but the distribution is now a quantised and more rectangular. The two examples are shown in Figures 5 and 6 contain 10,000 OpAmp datapoints.
Note: Statistical measurement error biases the shape of this graph, especially at very low offset value.
Appendix A - Primary configuration
Here is an example of a configuration bit-stream used as a primary configuration and where offset correction of the OpAmp in IO1 is required:
Line Hex Binary Notes
Particular points to notice are:
- Line 15, the auto-null clock is divided down by a user setting of 162, with the fixed 1000 of the associated pre-scaler this gives a 98Hz auto-null clock for the demo board 16MHz clock. NOTE: Anadigm Designer 2 Software will calculate this divider ratio automatically.
- Line 17, the IO1 OpAmp is selected for auto-null. Simply adding more bits to this byte and byte 15 would select other OpAmps for auto-null. NOTE: Anadigm Designer 2 Software will set these bits according to the user’s circuit.
- Line 28, the OpAmp has its inputs connected to the package pins (set by Software).
- Lines 29 and 30, the OpAmp is powered up in high power mode. If the OpAmp is not powered up (in either high or low power modes), it will simply ignore the auto-null cycle (set by Software)
Appendix B – Auto-null Reconfiguration
Here is the short auto-null reconfiguration. This byte stream is to start up auto-nulling following a primary configuration or a reconfiguration in which auto-nulling was disabled. Note that this byte stream will have no effect if the preceding configuration had auto-nulling enabled. This byte stream is designed to turn auto-null back on after it was disabled, and so force a new auto-nulling sequence to be initiated.
Hex Notes
D5 sync
-
ID
C1 Control byte
CB start with byte 11 (another block to follow) 00 bank 0
-
1 byte to follow
40 enables auto-nulling for the chopper amplifier 2A end of block
8F start with byte 15 (last block) 00 bank 0
-
2 bytes to follow
FF enables auto-nulling for all the CAB OpAmps 0F enables auto-nulling for all the IO OpAmps 2A end of block
00 dummy byte