I was told that it cost $10,000 for a HDMI license and another $5000 a year for the HDCP keys !! It may make a good kickstarter product in the future
I spent some time looking into this last night. I might be wrong but I think any HDMI license fees are included in the royalty that the chip manufacturer pays for the chip. In any case, you can purchase HDMI chips from several distributors. Checking further though I found Analog Devices will only sell you an HDMI receiver chip with HDCP enabled if you have an HDCP license. They also offer the chips with HDCP disabled. So if you can the chips you can use them and most likely they will work as they chip has no way of knowing if the license is paid up.
I also see that some HDMI repeaters and "Y" connector chips remove the HDCP content from the downstream leg, but I wouldn't count on this for future revisions. You could however do debug/development with one of these in the chain and a HDMI receiver with HDCP disabled, deferring the license cost until you were close to production.
But $5000/yr isn't all that much even in the self funded startup stage, if that is where you are. (been there and I can imagine you are already in much deeper than $5K) You can amortize it at $50 per unit on the first 100 units and probably charge $100 for it. You really just need to pay the license fee one year and buy enough chips to last for several years. If you have a friendly local distributor, he might even stock them for you and let you pay as you go.
In any case, plug and play with HDMI is what you need to get your product out of the DIY range and into higher volume markets where there is more volume and at higher price levels, so its worth some creativity in figuring out how to afford the investment.
I think what matters more than bit-depth or sampling rate or even DSD vs PCM is total signal to noise ratio at the analogue outputs. 192Khz does not mean much if any of total signal chains cannot handle it. For example, tweeter on LX521 is good enough up to 30kHz according to the manufacturer's specs. That is twice my maximum hearing frequency (about 16kHz). 96kHz is good enough - 4X oversample.
To correctly handle crossovers with FIR, FIR tap length is most important and then bit precision. MiniDSP's minisharc cannot handle crossovers for dipole woofer due to insufficient FIR taps.
If sampling / bit depth causes to degrade sound quality, you can apply convolution in computer at 64-bit double precision floating point and oversample at 384kHz and output to a multichannel DAC that support such format.
What I know is that I can hear the difference between 96khz and 192khz and DSD so. So there's more to this than theoretical frequency response of a given sample rate.
Personally, I wish they done an analog 4-way crossover.
So far as I know, the Nyquist Sampling rate theorem still holds and the Z transform still works for interpolating between samples. So long as the sampling rate is at or above the Nyquist rate, no information is lost. What over sampling does is move some of the quantization noise out of band and ease the anti-aliasing filter requirements post D to A. Correct me if I'm wrong and my education in this stuff took place decades ago but most of the benefits of over sampling could be obtained doing the bulk of the computation at the Nyquist rate and converting to a higher rate just prior to the DtoA step.
On the low frequency channels you could decimate downwards from 192K to 48 or even 24 or 12K !! Then when you are finished processing you interpolate back up to 192K for the DAC's. This way you save a lot of wasted DSP clock cycles for virtually no change in performance.
cheers
Yes I saw that from ADI but without the HDCP keys the equipment you are connected to will not talk to your HDMI receiver so I think that option is for setups where copy protection is not required.
cheers
I calculated FIR linear crossovers for LX521 (without any other EQs applied) and trucate them to 24bit/192kHz limit. All internal computations are done in 64bit floating point. Below are number of FIR taps;
Woofer: 11713
Mid low: 9997
Mid high: 4903
Tweeter: 159
Total: 26772 per channel / 53544 for stereo input
If output bit resolution is increased to 32 bit, then required FIR tap lengths are;
Woofer: 15003
Mid low: 14087
Mid high: 9281
Tweeter: 225
Total: 38596 per channel / 77192 for stereo input
I hope this helps.
Last edited:
Last edited:
Hi Tranquility Bass
How are you planning on implementing the decimation, more precisely the antialiasing filter?
Have you read the 2007 AES paper from four audio? They describe the decimation process used in the HD2, which is very clever but requires compensation in the filter target response (shallow low pass using an Albrecht window). It requires fewer taps and has less potentially adverts effects than a brickwall filter.
I am currently (in the process of...) implementing this compensation in rephase, so if you use the same low pass filters that would make it easier to do something a little more generic.
As far I know it's not a dead horse. It's a theorem not just a theory which essentially means it's mathematically undeniable. There is some benefit theoretically from a higher precision for the sample levels to higher bit rates but not for sample rates
I don't know anyone that actually implements full 4-way linear crossovers for LX521 with bit-perfect precision. I used a public information to understand the crossovers for LX521.
Regardless of sampling frequency choice, it takes 61msec of convolution to construct a second-order L-R low pass filter with linear phase and still bit-perfect to 24 bit.
Using lower sampling frequency for woofer low-pass filter sounds attractive. However, in order to avoid aliasing issues (i.e. creating high frequency artifact), another FIR filter needs to be applied as a part of upsampling process. I calculate a minimum length of FIR filter to meet bit-perfect accuracy. It is 291 in order to upsample 96kHz woofer crossover to 192kHz at 24 bit perfect.
I don't know if Sharc chip can apply two passes of FIR filters to implement lower sampling frequency for woofer and upsampling for output.
Bit-perfection takes a lot of DSP resources, indeed.
Last edited:
I learned about crossover settings based on LX521 Description .
Bit-perfect means that output samples are precise to 24bit integer using 64bit floating point internl calculation and any abosolute values less than 0.5 will be truncated to zero and thus determine FIR length.
In the same token, any DSP output values less than DAC's noise flow are effectively zero. That means a low quality DAC does not need longer FIR taps than a high quality DAC such as ES9018S.
The DSP challenge of the Linkwitz site is interesting. Someone had done a three way system with a passive crossover for upper and lower midrange which was similar to their ASP. I think the MiniDSP guy is now doing a program for 4 way DSP at 96khz. I wonder if it exceeds the resolution of the ASP though?
DSP_challenge
Steve Devino of Granite Rocks told me the current Metric Halo DSP is probably not up to the task at 192khz but their forthcoming product should be. BTW he's pretty good at programming DSP for audio.
DSP_challenge
Steve Devino of Granite Rocks told me the current Metric Halo DSP is probably not up to the task at 192khz but their forthcoming product should be. BTW he's pretty good at programming DSP for audio.