With the 8970A back working, what would be the first thing to do with it – well, let’s measure some gains and noise figures. Unfortunately, the 8970A alone won’t be sufficient, because it uses a small, external noise source unit, commonly refered to as a 346A, B (or C model, if you need noise up to 26.5 GHz). These sources are still widely used, although Keysight has introduced a new series, the N4000 series, but still the 346 models are very common, and available – this product has been around for 30+ years, not bad. The only downside – most of them seem to get lost or damaged, so they are rare on the second hand market, at least, if you don’t want to pay more then USD 0.5k for a used, out of cal, and scratched item, for a device that sells for USD 2.5k brand new.

Key characteristics of a noise source for noise figure measurements, and related tasks:

(1) The connector, preferably, get a 3.5 mm APC, then you can add a connector saver, and most of the small devices being characterized are SMA or 3.5 mm design; sure, have a few adapters at hand, or a SMA to N cable. A noise source with N connector is more sturdy, but also these connectors wear out, and aren’t all that hand except for directly connecting the noise source to the analyzer, which is not often done. Typically, the device-under-test (DUT) is connected with some short test cables anyway, and for calibration, you just remove the DUT, rather than all the cables.

(2) Flatness. The noise output needs to be so-called white noise, absolutely flat with frequency.

(3) Related to flatness, very low SWR. The various common DUTs, amplifier, mixers tend to have not too good SWR, so at least the noise source needs to have low SWR, otherwise, measurement errors will be enormous. Also, the SWR needs to be close, or the same, irrespective of the on or off state of the noise source.

(4) Well-known absolute noise power, measured in ENR, which is noise above a 290 K floor, -174 dBm/Hz (a 1 Hz bandwidth power density). 290 K is the Kelvin temperature of an average antenna on the surface of an average place on earth. Well, where are these average places that are at these constant 290 K…

(5) The driver input, commonly, a BNC connector that is driven by a 28 V DC signal. Most sources adopt this style of input.

This is one of these desirable items, in the typical used condition. Very similar device are available from Anritsu, NoiseCom, and others. The 346B has 15 ENR output, which is a good amount for general purpose application, maybe a bit too much for certain GaAs preamps, or other low level low noise applications; then you can just add a good (really low SWR) 10 dB attenuator.

…unfortunatly, I currently don’t own any of these extremely broad-band calibrated and well-working sources, and need to deal with less fancy apparatus, but let’s at least investigate what it is all about.

The block diagram (taken from the April 1983 issue, of the HP Journal, http://www.hpl.hp.com/hpjournal/pdfs/IssuePDFs/1983-04.pdf, shows the internal construction, still looking for a schematic of the current source, it seems to convert the positive 28 V signal, to a negative current, looking at the polarity of the noise diode. Maybe more about this later; to get proper accuracy and repeatability, it is a must to have a very constant bias current supply, on the order of 8 to 10 mA. It should provide a low noise DC current, without any large buffer caps, because the 8970A will switch it on and off periodically, to do the actual noise figure measurement. But there plenty of circuits around to accomplish this.

Most interesting, the matching network. Noise diodes have about 15-30 Ohms impedance, so this all makes sense. The strange stub is one of the secrets (the major secret) that ensures the 18 GHz flat output. The 6 dB attenuator improves the output SWR and SWR change from on to off condition. In fact, it is a good idea for any noise source design to have a high quality attenuator at its output, with at least 6 dB, or a bit more.

To replicate the 346B design, or at least a similar design that is good to a few GHz will remain a venture for future cold winters (good designs have been published by others but they all appear to lack flatness, and some use pretty costly noise diodes, and all need calibration that is not easily achieved unless you have access to a calibrated source).

For work demanding less accuracy, many design are pretty suitable to get reasonably flat noise of the desired power, in the 10s to 100s of MHz range.
This is one of the circuits that I have successfully employed.

It uses the breakdown of the emitter-base diode, according to the datasheet, about 5 to 6 Volts for any common NPN transistor (minimum values, actual breakdown might not occur up to 8-10 Volts). The noise source currently in use has a BC238B transistor, because it was the first one to grabbed from the junk box. Others will work as well, including BC107, BC548, 2N2222, 2N2904, and so on. The latter two appear to have a higher breakdown voltage. Obiously, there is no bias current regulator, and the 5k6 resistor will need to be adjusted to get the right level and flatness of noise in the desired range of frequencies. Sure, better results can result from a RF transistor like the BFR93, or other 4 GHz, or even 22 GHz type transistors – will give it a try back home in Germany because it really only makes sense in a proper RF setup, and on a small test board.

A quick test, to determine the gain and noise figure of a 6 dB attenuator. Attenuators have negative gain, equal to their attenuation value, and increase the noise figure of a system by the same magnitude.

For the time being, let’s call it close enough. With the simple noise source, calibration works perfectly fine from well below 100 MHz, to above 1 GHz, making it suitable for various general purpose application.