And what are the components you suggest to work on to boost the SNR in table top systems (generation, pump stability, electric noise immunity, electro-optical sampling vs antenna etc.)?
The answer may also depend on your application as well as your working definition of SNR. There are a number of definitions of SNR used in the THz community, e.g. some in the time-domain and others in the frequency domain, some for THz field others for power. I find the most useful is the dynamic range defined in Jepsen et al. Opt. Lett. v30 p29 2005, which is in the frequency domain for the THz field. With such a definition, it is typical to have a peak THz amplitude that is 3 orders of magnitude above the noise floor, while 4 orders would be extremely good SNR.
There's a few studies that compare photoconductive dipole antennae vs optical rectification & free-space electro-optic sampling (probably in ZnTe) - if i recall they can have roughly similar SNR. Typically PDA can win out due to the higher modulation frequency that can be used, as PDA are often used with oscillators while ZnTe is often used with amplified systems. See e.g. Cai et al. APL v73 p444 1998. Using ZnTe or other nonlinear optical methods can often lead to larger bandwidth than PDA, unless the feature size on the antenna is quite small, which can limit SNR. Generally broader bandwidth systems will have smaller SNR.
When using NLO methods, often a source a reduced SNR is competition between OR and unwanted two-photon absorption, see e.g. Harrel et al JAP v107 p033526 2010. So it might be best to have the OR THz generation occur with a collimated high-power beam.
Air-plasma generation, while broadband, can also be temperamental and quite sensitive to pointing changes, power fluctuations, pulse width changes, air-currents in the lab, etc.
In my lab, using PC emission and ZnTe for detection, the best I've achieved is S/N=5000 1/(Hz)^0.5 with an oscillator (10nJ/pulse) system, and S/N=2500 is typical. The noise floor is typically set by fluctuations from the laser, but on a good day, it's within a factor of 2 of shot-noise limit.
For typical linear measurements (transmission, reflection) this is not so important, because artifacts introduced by switching between sample and reference limit the minimum practical transmission change I can resolve to about 1%.
In optical pump/THz probe the signal to noise ratio is important, and limits the smallest transmission change I can resolve to about 0.1% without heroic efforts.
Just FYI, you can obtain much higher DR with a CW system based on harmonic multiplication of stabilized microwave sources, up to about 1.4 THz. The ABmillimetre VNA produces about 10 microwatts at 1 THz. Not much help if you've already invested in a time-domain system, I understand.
Jean-Paul, you mean you get a DR of 10000 at 1 THz? The ABM system will give 80 db DR over the 600-1000 GHz band. So your 40 db pulsed vs a CW measurement of 80 db. About what I would expect.