Phil Hobbs wrote: : I need a gizmo for making accurate measurements of low noise : temperatures from near DC to a few hundred megahertz. I'd like to be in : +- 0.1 dB territory for noise temperatures from about 25 K to 300 K. : Something pretty well based on first principles that doesn't itself need : calibration would be the ticket.
: I've been looking at switching between hot and cold resistors using a : coax relay, but they all seem to have return loss funnies down at the : 0.3 dB level.
: I'm leaning towards just putting an 0603 resistor on the end of a few : inches of 0.81 mm coax and silver-epoxying the shield to a small : Peltier. The heat leak won't be much (only about 0.6 mW/K for 10 cm : length. With appropriately placed Styrofoam, I can run that from : probably -30 C to +60 C, which will give me about 1.4 dB change in the : thermal noise. The measurement would basically be to run a few : temperature cycles, measuring the output noise as it goes.
: The output noise will be proportional to the sum of the noise : temperatures of the amplifier and the resistor. As long as the gain of : the amp stays constant, I can calculate its noise temperature as follows
: R = (measured total power at Th)/(measured total power at Tc)
: (Th - R*Tc) : Tn = ----------- : R-1
: Assuming the amp and power meter are linear, all the level calibrations : cancel out, so I ought to be sensitive only to errors in Th, Tc, and R. : Getting a 2% measurement down at 25K from this setup will require : something like 0.2 K temperature accuracy, which isn't that easy to do, : but I can probably do nearly that well with a diode-connected : transistor. Down at the 25K end, 5% wouldn't be that awful as long as : it sits still, and I can do that well with a YSI glass bead thermistor.
: Any relevant experiences/suggestions/criticisms welcomed.
I don't have much experience on ultra-stable gain contraptions, so I'd rather suggest ways to reduce the acuracy requirement.
If you can get hold of liquid nitrogen, through a colleaque in a local physics lab for instance, matters would become rather easy. LN can be carried around in a stainless steel thermobottle which you can get from a local camping store. The cold resitor can just be hanging from a twisted pair or a small-diameter coax. I have wired a full LHe dipstick by using prefabricated U.FL-U.FL cables (like DigiKey H11554-ND), which saved me the trouble of soldering the connectors to the cable. The SMD resistors specified at 25 ppm/C, or even at 50 ppm/C (I like Susumu RR0816Q series) hold their values quite well down to LHe, but of course this is something you may want to calibrate out, depending on your accuracy requirements.
The role of the hot resistor in the cold/hot resistor noise figure measurement is just to calibrate the gain of your amplifier chain. As John L. pointed out, it is much easier just to use a cold *attenuator*, and "warm it up" by feeding additional noise into it. At frequencies I'm typically interested in I can do it with an ARB generator. At microwaves, where you usually must calibrate everything, the hot resistor *is* nice, as a sure-fire self-calibrated flatband noise source. With the attenuator, pseudorandom noise is just one convenient way to do the gain and frequency response calibration (when doing noise measurements you have a spectrum analyzer hooked at the amp output anyway), but a chirp or a sine sweep would work equally well.
If you are not operating at very high frequencies, it is possible to just measure separately the voltage noise of the amplifier by using a room-temperature source resistance significantly lower than the noise match, and current noise by using a source significantly higher than the noise match. Make the load resistors into the attenuator shape so that you can measure the gain on-the-fly. You then calculate Tn = un in / 2 kB . If the source is ac coupled it does not affect the amplifier biasing as it otherwise would. Parasitics are likely to cause a frequency rolloff when the source is away from the designed match, but you can recognize that (and to an extent cancel out) from the measured gain response. But a few hundred MHz may be difficult this way.
One more possibility is to make an actively cooled resistor out of an amplifier, following Percival's ideas, but to make it unconditionally stable and figure out *its* characteristics may be more work than your attempt for ultra-stable gains. I think I have seen this suggested somewhere ... ahh, here it is: Frater and Williams, IEEE Tran. MTT, vol 20 no 4 p 344,
1981.
Regards, Mikko