How "high" is the level? Depending on the bandwidth and power level, you may want to avoid this method on some devices. Passive devices and amplifiers through VHF should be no problem. In the microwave range, some amplifiers can be damaged by noise testing - they can't handle the peak voltages, even though the average power may be quite small.
The fastest, tiniest devices are most at risk because they can only take
so much peak energy, regardless of the power spectrum. To get high
average power in broadband noise signals, large voltage excursions are
needed. You can see the envelope of these in the time domain, and
consider what effect the peak voltage may have on the DUT and the measuring equipment. If the peaks are equivalent to less than the peak CW voltage that the device can handle, then it should be no problem.
Also consider the same regarding the SA input capability. This is aggravated by the low test signal power spectral density - in order to see the fairly small resulting power levels, the front-end needs to be set to higher sensitivity ranges, so the front step attenuator that usually protects somewhat against typical RF CW levels may be set to minimal attenuation. It's good practice to keep at least 10 dB attenuation up front, but it's easy to forget during operation, exposing the mixer to possible damage.
If the SA goes to microwave ranges, but the required DUT and noise bandwidth are much smaller, say audio up to VHF, I'd recommend adding a diode-clamp type limiter to protect the SA input, just in case. Then you won't have to worry about the peaks, and can try to get as much noise source power as possible - you'll want all you can get to provide reasonable test levels.
For lower frequency SAs and such with high impedance input, there's usually plenty of input protection, so noise testing should be no problem.
Ed
Dave McGuire