Hi Fons,
Thank you for your comprehensive response.
On 13/12/11 22:31, Fons Adriaensen wrote:
On Tue, Dec 13, 2011 at 09:31:27PM +0000, Roger James wrote:
My current decoding workflow is as follows.
1. Recover B format from UHJ using convolvers and the impulse responses
provided by Angelo Farina.
You could also use jconvolver / fconvolver. An UHJ decoder preset
is provided.
I used jconvolver on my last experiments with good results.
2. Pass the recovered B format into Fons Adriaensen's Ambdec decoder
using his ITU speaker layout configs.
3. Encode the resulting speaker feed outputs into AC3 using "aften".
My questions are.
1. Should I experiment with shelf filters between steps 1 and 2, or does
the Ambdec decoder already do that internally (I saw a reference to it
doing this somewhere).
If you are using 'itu5.1-ord1-optim.ambdec', you have the
equivalent of a decoder with shelf filters. The actual
implementation is different, but the result is the same.
I thought so. Thanks for confirming that.
2. If the Ambdec decoder does this how should I configure it to account
for the original source material being UHJ?
I don't have any information as to what would be required.
OTOH I suspect the differences will be small. Also any
special 'UHJ' shelf filter, if you can find it, has probably
been computed for some regular layout (e.g. a square).
In this case the layout is irregular, and the shelf filters
are already optimised for that layout. Just adding those
for UHJ on a regular layout will not produced the correct
result. The information you need is the *difference* between
shelf filterers for standard B-format and those for UHJ.
Yes, the "difference" is the nub of the matter! The only data I found
was the graphs and diagrams for the "Integrex" hardware decoder, Table
3.2 and Figure 3.3 in the http://ambisonic.info/info/ricardo/shelfs.html
reference. These are difficult to read but assuming that HF is to the
right appear to show that the UHJ W filter has about 3dB more HF gain
and a more curved shelf than the B W filter. The UHJ XY filter looks the
same apart from having a "straighter" shelf.
If you look in the preset file, there are two matrices
defined, one for LF and one for HF. There are also
'order_gain' values for each matrix. Only the first two
of these matter (the others are for higher order). The
first is applied to the W column of the matrix (the first
column), the second to X and Y (the second and third column).
By changing these 'order_gain' values you can obtain the same
effect as would be provided by additional or modified shelf
filters. But I have no idea which changes would be optimal for
a B-format input derived from UHJ.
I guess that these values have no influence on the "straightness" of the
shelf, but that lifting the HF gain of the UHJ W filter might be an
interesting experiment. But this really depends on whether I have
interpreted the diagram correctly :-) (and I suppose on how the Integrex
decoder performed).
With the values defined in the preset you get a 'systematic'
decode at LF, and one that hase been optimised for rE magnitude
and direction accuracy at HF.
Ciao,
Roger
_______________________________________________
Sursound mailing list
Sursound@music.vt.edu
https://mail.music.vt.edu/mailman/listinfo/sursound
