Noise of a current source

Precision current sources are critical electronics in atomic physics. They can serve as power supply for diode lasers or drive a coil to generate a precise magnetic field. We sometimes need to have the load grounded in experiments to prevent short-circuit on the conductive optical table. This demand is often achieved by using an instrumentation amplifier(INA). The current flow through a sensing resistor and the voltage drop on that resistor is sampled by INA and feedback to the output stage. We find AD8429 from Analog Devices a great INA choice for a bi-polar current source with the input noise density as low as 1nV/\sqrt{Hz}.

When facing with extremely noise sensitive application, we should have a second look at our design. In our old design, the Gain=1 is set without gain setting resistor attached. This seemed very convenient because we do not want gain drift. The mismatch of the temperature coefficient of the on-chip resistors and the external gain setting resistor will sure cause terrible gain drift. In fact, the total input referred noise of an INA should include the output noise part attenuated from the feedback network. The AD8429 has a very decent input noise density 1nV/\sqrt{Hz}, but a relative higher output noise density 45nV/\sqrt{Hz}. When the INA runs at Gain=1, the total input referred noise is dominated by the unattenuated output noise of the chip. The result is nearly  45nV/\sqrt{Hz} noise density on the input after RMS summing of the two kinds of noise!

An easy way to fix this is to increase the gain of the INA. In our configuration, a 200m\Omega current sensing resistor is used under the maximum current 5A(1V voltage drop). If we can amplifier that to 10V, which is still inside the rail of the INA, the output noise contribution will also be greatly reduced to 1/10. However, it is difficult to just place a resistor on that chip due to the temperature drift mentioned before. Instead, I find AD8228, which has a pair of internal matched gain setting resistors, providing Gain=10 and Gain=100 through simply opening or shorting the gain setting pins. For its fixed gain, the noise density of AD8228 referred in the datasheet includes the output noise contribution on the input noise. This means that the 15nV/\sqrt{Hz} should be the final input referred noise density of AD8228 under Gain=10 setting. Definitely much better than 45nV/\sqrt{Hz}!

Lock laser to an atomic transition without modulation

There is always possibility that you need to stabilize a laser but do not have an appropriate modulator like AOM or EOM in hand. The dichroic atomic vapor laser lock(DAVLL) technique provides a really simple and robust modulation-free way to do that.

Although there is no modulation in DAVLL, the error signal of it is still resemble to the FM spectroscopy and the prevailing Pound-Drever-Hall(PDH) locking technique with a dispersive shape, which provide a stable zero-crossing point for locking. The dispersive error signal comes from the differential measurement of the Zeeman split levels.

The apparatus lack the complex of difficult alignment of the general modulators. Since the polarization in this technique is really important, good quality polarization beam splitter(PBS) with correct coating should be used.


The photo of the apparatus.


The coil and the heating strip.


With the pump beam(brown) blocked, and only observing the signal PD, we get the typical doppler broadened absorption signal.


With the pump beam unblocked, we can clearly see the hyperfine features in the saturated absorption (doppler free) signal from signal PD.


When we turn on the balance mode, observing signal PD with reference PD subtracted, we can see the dispersive shaped DAVLL error signal. We can also easily recognize the hyperfine transitions as well as crossovers from this snapshot.


And the last step is locking. The residue noise of after locking is shown below. This apparatus achieves sub MHz frequency noise according to a WS-7 wavemeter and will not easily get out of lock(as long as the laser is not drifting too far away).


Thanks Ben and Loic for kindly helping me debugging the system.