Advertisement

Blog The Signal Sped Up

Designing higher frequency active filters to drive differential input high-speed ADCs

With newer precision Fully Differential Amplifier’s (FDA’s) providing more than 800Mhz Gain Bandwidth Product (GBW), the available frequency range for active filter designs combined into the last stage interface to an ADC have moved beyond what legacy literature would suggest. Steadily improving amplifier choices, and RC value adjustment routines for GBW, have increased the frequency application range for these active filters. A recent >30MHz active filter design will be described here and updated to more accurately fit the desired response shape.

Example design driving a low power, quad channel, 12-bit, 50MSPS ADC

A complete design incorporating a JFET input stage into a 4th order single to differential stage using the THS4541 (1) FDA can be found in a recent (2016) reference design (2). One final iteration (section 8, ref. 2) added another pair of poles into the FDA stage as a Multiple FeedBack (MFB) active filter prior to a passive 2nd order RLC filter at the ADC inputs. These two stages of 2nd order filters were intended to implement a 4th order Bessel filter at 20Mhz (F-3dB ). One source for the filter targets can be found in the Intersil filter designer (3). Here, we only need the Fo and Q for each of the 2nd order filter stages as shown in Figure 1. This tool appears to be the only extant vendor tool supporting frequencies this high in an active filter implementation, although there are no FDA’s in this (or any) vendor tool.

Figure 1

Filter stage targets from the Intersil Active Filter Designer (3).

Filter stage targets from the Intersil Active Filter Designer (3).

The first pair of poles were easily added to the THS4541 FDA single to differential stage as an MFB filter design. The 2nd pair are implemented as a passive differential RLC filter at the ADC inputs. Figure 2 shows the TINA (4) simulation circuit for this original design. At the time of this design, a standard RC value selection was not employed – that step will be added here as well.

Figure 2

Click here for larger image 
Original 4th order Bessel design simulated to ADC inputs (ref 2, page20)

Original 4th order Bessel design simulated to ADC inputs (ref 2, page20)

This overall low pass filter was intended to be DC attenuating with 20Mhz F-3dB . The simulation shows this to be slightly off at about 20.9Mhz. Bench testing this single ended JFET input to differential ADC driver with this 4th order filter showed the excellent performance of Figure 3.

Figure 3

Click here for larger image 
Full scale 5MHz input FFT for original circuit + ADC showing 69.9dB SNR and 82dB THD.

Full scale 5MHz input FFT for original circuit + ADC showing 69.9dB SNR and 82dB THD.

Testing the response fit for the active filter stage.

The small signal response for the active filter stage can be simulated and compared to the ideal target. One nuance moving the response off target is the differential input capacitance (C2, Figure 2) used to improve the phase margin in this design. The THS4541 is not unity gain stable (Figure 1, Reference 1) and this capacitor transitions the AC noise gain to a higher level at loop gain x-over (Reference 5). The high frequency noise gain in Figure 2 is set by the input differential capacitance (doubled to make it single ended) divided by the feedback capacitance (1+(2*5.2pF)/3.6pF = 3.9V/V). This higher noise gain (to improve phase margin) effectively reduces the 850MHz GBW of the THS4541 to 218Mhz in a 31Mhz active filter implementation. This modest GBW margin would definitely suggest that a GBW RC adjusted flow might be useful. Measuring the simulated response at the output of the FDA in Figure 2 (keeping the RLC filter load in place) and comparing to the ideal response gives Figure 4.

Figure 4

FDA filter stage response vs ideal target.

FDA filter stage response vs ideal target.

The higher peaking (Q) than targeted can explain the higher overall F-3dB bandwidth than intended. This original design was producing an F-3dB at 36.6Mhz vs the intended 35.7Mhz. Extracting the simulated response details and comparing to the intended design shows the large fit errors in Figure 5.

Figure 5

Click here for larger image 
Extracted closeness of fit on Fo and Q.

Extracted closeness of fit on Fo and Q.

Improving the MFB stage design adjusting the RC values for GBW

The design of Figure 2 was done without applying a more sophisticated GBW adjust flow on the RC values. This flow, using a cubic pole expression including the amplifier GBW (Reference 6), can be effectively applied here to fine tune the RC values possibly improving the fit to target. As noted in (6), including an op amp summing junction noise gain tuning capacitor in the cubic polynomial coefficients is possible without changing the order of the transfer function. That would be necessary here to apply a non-unity gain stable FDA. Figure 6 shows the slightly modified RC values that will give a much closer fit in this relatively high frequency design.

Figure 6

Click here for larger image 
 Modified RC values in the THS4541 active filter stage to adjust for GBW and C2.

Modified RC values in the THS4541 active filter stage to adjust for GBW and C2.

Putting these RC values into an ideal op amp 2nd order MFB response extraction will appear to be mistargeted. However, Figure 7 shows these are adjusting very successfully for the THS4541 850Mhz GBW and the 5.2pF C2 across the FDA input pins.

Figure 7

Updated MFB filter design response vs ideal.

Updated MFB filter design response vs ideal.

The best fit RC standard value algorithm used here focused only on Fo and Q fit and allows the DC gain to go off 1 – E96 resistor step value if that improves the filter fit. The DC gain can usually be adjusted back to target in another stage – like the input JFET stage used in this reference design (2). Extracting the closeness of fit parameters shows the improvements in Figure 8 (compare to Figure 5). The 35.69MHz = F-3dB has moved much closer to the intended 35.73MHz.

Figure 8

Click here for larger image 
Fit errors using a GBW adjusted RC design flow.

Fit errors using a GBW adjusted RC design flow.

To complete this redesign for closer fit, also adjust the RLC filter to standard values and back out the ADC input capacitance from the final differential C – taking it down from the exact 80pF solution to the next 75pF standard value. Simulating this completed redesign shows the much closer nominal fit at 20.35Mhz vs the targeted 20Mhz F-3dB only adjusting RC values in this 4th order active + passive filter solution.

Figure 9

Click here for larger image 
Updated design with standard values and improved design fit.

Updated design with standard values and improved design fit.

The example RC values developed here should also work well with the similar ADA4932-1 from ADI (8).

Modern high-speed voltage feedback op amps and FDA’s can deliver higher frequency active filter solutions than commonly assumed. The only tool currently supporting a GBW adjusted flow appears to be the ADI Filter Wizard (7). Some of the features that appear to be missing from all the current vendor tools include the following:

  1. No attenuating MFB filter designs are allowed – but physically easy to implement.
  2. FDAs are not directly supported (FilterPro (9) does show designs, but only for an ideal FDA)
  3. Noise gain tuning with a capacitor on the summing junction(s) for phase margin improvement is not directly supported. It can be added as a final step, but that will interact with the response shape if not initially included in the GBW RC adjusted value flow.
  4. The allowed design frequencies are lower than possible with today’s higher speed devices. The ADI Filter Wizard delivers buffered RLC designs if asked for this solution. TI’s Webench Active Filter Designer (10) constrains frequency entries to a maximum 10MHz.

Modifying channel 3 in the reference design (2) to the improved RC values shown here would slightly improve the step response as the initial design was overshooting more than expected out of the active filter stage.

Even though these new RC values provide a slight improvement, they will deliver a much better nominal fit to target where the next step would be to run RC tolerance Monte-Carlo to get the response spread around this more centered nominal.  The ADI tool (7) also supports this feature, but not with FDAs nor in this frequency range. Lower frequency designs using an op amp of the same GBW as the intended FDA can certainly take advantage of the GBW adjust and monte-carlo spread features in the ADI tool. The op amp design could then be easily adapted to an FDA version using a device with similar GBW.

Improved THS4541 active filter references

  1. TI, THS4541, Negative Rail Input, Rail-to-Rail Output, Precision, 850Mhz Fully Differential Amplifier>.
  2. TI Designs TIDA-00799, Quad-Channel, 12-bit, 50-MSPS ADC Reference Design with Low-Noise, Low-Distortion, DC and AC Inputs
  3. Entry page for the Intersil online op amp design tools. Login required.
  4. TINA simulator available from DesignSoft for <$350 for the Basic Plus edition. Includes a wide range of vendor op amps and FDA’s and is the standard platform for TI op amp models.
  5. Design Methodology for MFB Filters in ADC Interface Applications, Michael Steffes, Feb. 2006
  6. Planet Analog, Feb. 2018, Michael Steffes, Include the op amp gain bandwidth product in the Rauch low-pass active filter performance equations
  7. Entry page for the Analog Devices online Active Filter Wizard. Login required.
  8. ADI, ADA4932-1, Low Power, Differential ADC Driver
  9. FilterPro users guide updated 2011.
  10. Entry page for the Texas Instruments Webench Filter Designer. Login required.

0 comments on “Designing higher frequency active filters to drive differential input high-speed ADCs

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.