Oct 4, 2011

In this post we investigate different configurations of the lead-lag filter, to determine what the best setup for the Rev4 squid controller boards is for deployment in SPTpol and PB.

Set up for these tests

• EBEX wafer 410-07
• 6 combs with ~12-16 detectors each, spread over 300 kHz - 1.2 MHz
• 2x Blue DfMUX boards
• 4x rev3 red mezzanine boards
• EBEX-style superconducting stripline, without a pigtail
• McGill cryoelectronics board controlling the fridge and thermometry. Bolos held at ~ 750 mK.
• DAN firmware

The lead lag filter is formed by the squid dynamic output resistance Z_SQUID ~ 100 ohms, the lead lag capacitor C_LL ~ 1nF, and the lead lag resistor R_LL.

The lead-lag filter turns on at frequency:

1. /(2pi Z_SQUID C_LL ) ~= 1.6 MHz

and turns off at frequency:

1/(2pi R_LL C_LL) ~= 8 MHz for R_LL = 20 ohms.

and attenuates by a factor

R_LL / (R_LL + Z_SQUID)

Rev 3 - Original SPT-SZ style lead-lag

Figure1 shows a nuller network analysis for the rev 3 board.

On this board the lead-lag is formed between the ~15 ohm Manganin wires in the 4-point connection and the 1nF capacitor at location Cx17.

Figure 1.

This configuration has been in operation on SPT-SZ and APEX-SZ for years - it works, but the configuration isn't optimal, since the hump at 4 MHz is caused by the phase shifts in the lead-lag. In this region, we see a lot of ADC rails, and noise from elsewhere (e.g. the cryoelectronics boards) are amplified. We want to improve on this, and remove the strong dependence on small variations in the manganin wire impedance.

Rev 4

Figure 2. shows the schematic for the rev 4 SQUID controller board. For the rev 4 tests the old lead lag has been removed ( Cx17 and Cx20 ) and the new lead lag has been altered ( Rx48 and Cx04 )

Figure 2

Version 1 Cx04 changed to 1 nF, Rx48 changed to 10 ohms. Turn on predicted at 1.6 MHz, turn off at 16 MHz. Figure 3 shows the resulting nuller network analysis.

---

Figure 3: C_LL = 1nF, R_LL = 10 ohms

whoa - ouch - this one is no good. It causes phase shifts from the lead lag creating an instability that is similar to the one we had in Rev 3. This instability is impinging on our bolometer bandwidth, so we reject it (strongly).

Version 2. Cx04 changed to 1 nF, Rx48 changed to 20 ohms. Turn on predicted at 1.6 MHz, turn off at 8 MHz. Figure 4 shows the resulting nuller network analysis.

---

Figure 4: C_LL = 1nF, R_LL = 20 ohms

ahhh, nice!

Here the turn-off of the lead-lag has been moved to lower frequency, handling the phase shifts. The instability is pushed up to 7 MHz, and is much smaller in amplitude. This will reduce the ADC overloads, and move these features further from our bolometer bandwidth.

This is the lead lag configuration we choose.

Version 3 Cx04 changed to 1nF, Rx48 changed to 50 ohms. Turn on predicted at 1.6 MHz, turn off at 3.2 MHz. Figure 5 shows the resulting nuller network analysis.

---

Figure 5: C_LL = 1nF, R_LL = 50 ohms

This is the version chosen by applying "Martin's ansatz" from his thesis. It chooses the best choices or the turn on and turn off frequency, and accepts whatever attenuation comes from that, which in this case is a much smaller 1/3 attenuation.

Nasty - the instability has grown huge - and is probably peaking out of band. This will lead to a large number of ADC overloads.

CONCLUSION: these measurements suggest the best choice for the lead-lag is 1nF, 20 ohms.

-- JamesKennedy - 2011-10-04 * rev4_lead_lag.png:

---

This topic: AnalogFMux > WebHome > SquidControllerRev4Mods > Lead_lag_modifications Topic revision: r2 - 2011-10-04 - MattDobbs

---