As always, I’m open to suggestions as to how to improve this test or reasons why my test is not valid – let the conversation begin!
To get back to the actual topic, this is why the test isn't valid. I'm going to use the illustrations from the SOS article (so acknowledgements to Hugh Robjohns and the SOS website for those) to show what is going on if both compressed and uncompressed channels are set to unity gain to avoid being skewed on the subject by variable gain relationships between the two signals and just talk about only about compression make-up gain from an identical starting point from both channels.
Remember that the original test didn't start from a perfect null anyway, and just to illustrate the small amount of deviation when no make up gain is applied to the compressed channel and the threshold is set to -23 db at a ratio of 50:1 we see this curve from the SOS article.
The straight diagonal plot is of course the unprocessed signal and the purple dotted line is showing the compressed channel, the green line is just demonstrating a compressor on an insert but it's the red line that is is starting to show 'what's really going on'. It's showing the 6 db of gain added as a result of adding the two signals together below the threshold as we would expect and it's showing a gentle deviation from the straight compressed insert which has 6 db of make-up gain applied.
This is where I believe the test in the RT article is invalid. The deviation is so slight that given a not so perfect null at the start you can achieve as bad a null as before just by altering the ratio on the non-parallel compressor used for the comparison. Not only that the article implies that all parallel compression does is equate to changing the ration on an inline compressor yet it doesn't show how you'd use a parallel setup to change ratio in any meaningful way, nor does it demonstrate at all anything of 'what is really going on'
So now we move on to adjusting the make-up gain on the parallel compressor and we see that the gentle curve when no make-up is applied start to accentuate into a form that more resembles upward compression and will also largely retain the transients.
So this graph shows what happens when extra make-up gain is applied to the compressed channel, set with a threshold of -40db and a ratio of 50:1, this plot was originally done with extra identical parallel compressors, the green lines (which was later found to be unnecessary as adjusting the make-up gain on a single instance produced an identical result the dotted red lines).
The plots here show 3,6,9.5,12,14,15,17,18 db of make-up gain respectively.
Points to note are that whilst the noise floor is still going to be raised it will be considerably less than it would be using an insert compressor provided it is setup properly, there is a clear form of upward compression occuring even if not in the strict meaning of upward compression and most definitely the transient peaks are much less affected than they would be by using inline compression.
I would say however if it is true upward compression you are after the Selig's Leveller RE is excellent and I've certainly never found another device like it, but if you are a fan of the NY Compression sound there is certainly more going for it than a simple ratio adjustment.
I'd recommend using a specific compressor for it too, it seems to work best for me with a comp that doesn't mind beomg set up for a very low threshold and high ratio, so something 1176 ish but with a threshold control like the Softube FET (although I genarally use the Stilwell Rocket) is ideal, and don't forget that it is only above that threshold where the characteristic 'upward compression' transfer curve occurs. Personally the best result I've got out of using it is with pre-mixed drum material say where I want to protect a nice and delicate hi-hat pattern over monstrous pumping and breathing kicks and snares, but I'm sure many of you have found plenty of other uses for it as well.