[Radio Wrangler]

Noise figure has long been a rather dubious specification. Unless you get data that was similarly measured and done with great care, the comparisons are likely to be misleading. A number of radio equipment manufacturers, lacking the machinery to measure it properly, were inclined to just quote the spec of their front end device.

Quite often you got expressions like .... so many microvolts input to produce XdB S/N ratio without mentioning what the input impedance, or the bandwidth was.

Someone with a collection of tuners could measure them nowadays, of course, but there are still difficulties.

Precision noise sources are normally only calibrated from 10MHz upwards. Noise figure meters like the HP 8970A only work down to 10MHz, and worst of all, the whole thing assumes it's measurement bandwidth (4MHz) is narrower than the thing under test.

The later Agilent N897x family of noise figure analysers share the 10MHz low frequency limitation as far as their internal preamps go, but the measurement bandwidth can be narrowed. I can't remember how narrow Andrew set the limitation in software, but I have used prototypes down to 100kHz where the DSP card is fitted.

I'd use a 15-ish dB ENR noise source with a fixed 5dB pad and then a minimum loss 50 to 75 Ohm pad. Measured carefully, the pad losses can be taken into account.

The 15 dB ENR sources give a small variation in output impedance between noise-on and noise-off states. This can modulate both the gain and the noise figure of the device under test in a way that puts a significant error into the results for low-noise devices. The 6dB ENR family of noise sources simply include an attenuator to dilute the Z variation. You can DIY this by adding a precision attenuator to a 15dB ENR family member, and this is an opportunity for 50 to 75 Ohm matching.

David