Audiocontrol Hyperion Processor

[masterhw]

DSP latency isn’t just a factor of ‘horsepower,’ but is also related to the choice of filters. I know we’re used to faster chip = can do the ‘work’ in less time, but linear phase FIR especially needs a continuous buffer inherent to the design, which adds unavoidable latency.
Dirac and other DSP architectures use mixed phase and minimum phase FIRs to balance this latency, but it’s still not as simple as faster chip = lower latency.

If D&M have implemented Dirac ART competently on SHARC+ embedded chips, I would expect marginal improvement at best even from a supercomputer. Faster chip bringing better DSP is a powerful idea that marketing is probably happy to support, but that doesn’t necessarily make it reality.

 

[Highfidelity]

I know that. It’s not about how fast the chip is. It’s the GFLOPS of processing power. If the DSP is capable of more GFLOPS of DSP processing, it allows the processor manufacturer to take advantage of higher tap filters. And it’s because of the higher tap filter the latency is added. It’s not possible to increase the latency with higher tap filters if those filters would exceed the processing capability of the DSP.

For these reasons the 100ms max eARC latency compensation is never the limiting factor with AVP’s that have eARC. It’s always the DSP power limitation.

This is also why processors like the Trinnov which is capable of 32768 tap filters don’t bother with eARC. As eARC mode would severely handicap the capabilities.

 

[Steakhouse]

GFLOPS is a measure of how fast a chip is. Your sentences thus makes little sense.

 

[Highfidelity]

“Processing power” is often more accurate than simply calling it a measure of “speed,” because GFLOPS reflects only raw computational capacity and ignores critical real-world limitations like memory bandwidth, data movement overhead, architectural efficiencies (e.g., accelerators, SIMD, DMA), code optimization, interrupt latency, power consumption, and whether the workload is even floating-point dominant; as a result, actual sustained performance in practical applications can be significantly lower, making “DSP power” a broader, more holistic term for the processor’s overall capability to deliver reliable, real-time signal processing.

 

[PGAMiami]

Where did you read about those details? It’s possible the ARM chips are used for other things than DSP. The Trinnov CI also uses ARM chips and FPGA’s, but they’re used for other functions other than the DSP.

Regarding the ART algorithms I’ve never looked into the fine details of the algorithm. I prefer to manually tune rooms with my own DSP. And since I use 4th order active crossovers which put all the drivers back in phase, there’s very little to correct in my setups. Dirac is mainly a bandaid to correct flaws introduced by passive crossovers. I haven’t found any room correction to date that doesn’t degrade the sound more than improve it in my systems. The Trinnov on the other hand is an excellent measurement tool to quickly verify if time, phase, and frequency response is optimized in the system. So I plan to buy a CI just to use as a measurement tool. And manually fine tune in my own DSP until I’m able to achieve the Trinnov target curve without running the audio through the Trinnov correction.

Processing Power | Hyperion

Hyperion processors utilize precision digital signal processing that suppresses jitter-related time-domain errors, preserving phase alignment, full dynamic range, and accurate harmonic structure for lossless signal delivery.

www.hyperionprofessional.com

EFFICIENT PROCESSING OF HIGH COMPUTATIONAL DEMANDS​

The soul of a Hyperion surround sound processor is its System on a Chip (SoC) quad-core, ARM-based design. Hyperion engineers custom-authored the Linux-based operating system in tandem with the platform's evolution, heavily optimizing the software to utilize multiple processor cores effectively.
The SoC/ARM RISC (Reduced Instruction Set Computer) microarchitecture design emphasizes system speed processing with much lower power and thermal envelopes. Robust hardware acceleration supports specialized 1 GHz DSP chipsets for high-throughput performance with computational speed equivalent to X86 CPU-based proce

 

[Steakhouse]

That reads like a similar approach that Storm uses and afaik also Denon et al. An ARM CPU for the GUI and general tasks. DSPs for sound processing. Since we dont know how much of those DSPs and what kind of it is hard to say how fast they are.

 

[PGAMiami]

That reads like a similar approach that Storm uses and afaik also Denon et al. An ARM CPU for the GUI and general tasks. DSPs for sound processing. Since we dont know how much of those DSPs and what kind of it is hard to say how fast they are.

My point is that an x86 processor has the throughput to run 60,000 tap filters. Latency is another issue altogether. The sentence I’m referring to is about the DSP capabilities not general tasks or the GUI. So my question is why did they bother to have this much processing power? Are there higher tap or higher sampling rate versions of Dirac they may run on these DSPs vs the what’s on a basic AVR with less powerful chips? Is the basic D&M Dirac good enough? Is Storm’s Dirac any different or better? Can their DSP switch to higher latency filters when eARC is not in use? Is this for future DSPs? Roon Rock on a NUC with Linux operating system is a very capable DSP. Looks to me like the APR-16 has similar hardware capabilities.

 

[Highfidelity]

Yeah it likely uses the ARM for system management, GUI, networking etc. And dual ADSP-21569’s for the DSP role. As they’re the best 1ghz DSP’s available for the task.

 

[PGAMiami]

Yeah it likely uses the ARM for system management, GUI, networking etc. And dual ADSP-21569’s for the DSP role. As they’re the best 1ghz DSP’s available for the task.

 

[PGAMiami]

So very similar to Storm. Trinnov uses a CPU?

 

[Highfidelity]

The Trinnov is based on a PC motherboard and Intel CPU. And the Dante board is PCI-e based. Then they have an aux board for the HDMI ports, along with AES/EBU. They likely run a Xilinx Zynq hybrid Arm/FPGA to handle the computing needs on the aux board.

The Storm uses an ADSP-21598 main Atmos decoding DSP along with quad ADSP-21469’s for post-processing. It also uses a Raspberry pi with Arm chip to run the GUI.

 

[PGAMiami]

The Trinnov is based on a PC motherboard and Intel CPU. And the Dante board is PCI-e based. Then they have an aux board for the HDMI ports, along with AES/EBU. They likely run a Xilinx Zynq hybrid Arm/FPGA to handle the computing needs on the aux board.

The Storm uses an ADSP-21573 main Atmos decoding DSP along with quad ADSP-21469’s for post-processing. It also uses a Raspberry pi with Arm chip to run the GUI.

So then to summarize and simplify, the Trinnov is a different animal, essentially a prepackaged computer based product. The Storm and APR-16 are more like an AVR but with top of the line chips, and the most current of these chips are in the APR-16. The APR-16 hardware should be comparable if not superior to the Storm, if anything because it’s a newer design.

 

[Highfidelity]

The Storm has quite a bit more power available. The 21598 main DSP is capable of 24 GFLOPS on its own. The 4x 21469’s are a bit dated, but handle the task of post-processing on their own with 10.8 GFLOPS. So combined there’s 34.8 GFLOPS available. Where the APR-16 would only have a max of 16 GFLOPS if fully optimized with 2x 21569’s. But it also only has 16 channels to process. Where the Storm can do up to 32. So there shouldn’t be any issues at all for the APR-16 to handle the same level of ART as the Storm. Where lesser AVP’s would need a watered down version with less taps.

 

[Steakhouse]

My point is that an x86 processor has the throughput to run 60,000 tap filters. Latency is another issue altogether. The sentence I’m referring to is about the DSP capabilities not general tasks or the GUI. So my question is why did they bother to have this much processing power? Are there higher tap or higher sampling rate versions of Dirac they may run on these DSPs vs the what’s on a basic AVR with less powerful chips? Is the basic D&M Dirac good enough? Is Storm’s Dirac any different or better? Can their DSP switch to higher latency filters when eARC is not in use? Is this for future DSPs? Roon Rock on a NUC with Linux operating system is a very capable DSP. Looks to me like the APR-16 has similar hardware capabilities.

There is no single x86 CPU from which one could generalize performance. We dont know how many taps the CPUs in the Trinnov can handle in the end when you really want to push settings. We only know that their CPUs are much more powerful also for DSP workloads than those FPGA chips fron analog devices that Storm et al use.
Why do they have that much power? Because some middle class Intel CPU is actually not that expensive and you want the units to be future proof. Most of the engineering work is to code for DSP work and decoders and build the custom Linux platform. Once you have that you are much more flexible than with FPGA chips and also because you dont have to use hard wired decoders. That is also why Storm uses the ADEC board now. This is not a specific decoder board, it uses an ARM CPU for which the decoder software can be programmed. With that respect they leaned a bit to the Trinnov approach.

 

[welwynnick]

Well this was interesting. Hyperion originally claimed 130dB DNR / 120dB THDN for the APR-16, and I was sceptical about that.
They have clarified things, and now say this about the DACs, which appears to refer to the performance of the DAC chips themselves.

Hyperion engineers selected the Hyperstream IV ESS SABRE ES9039Q2M 32-bit, 2-channel audio D/A converters for their measured performance parameters, including an ultra-high dynamic range of +130dB and an extremely low total harmonic distortion plus noise (THD+N) of -120dB.

For the spec of the APR-16 unit itself, they now say:

0.0002% THD+N is -114dB. That's less than the chip spec, but better than any other AVP. Ever.

They also quote different SNR at different output voltage. First, because the DACs are most likely being used in current output mode (that's how high performance DACs perform best), so the DACs will feed I to V converters, and only after that can the performance be expressed relative to an output voltage. So that's a pretty good indication in itself that those aren't chip specs.

Second, the SNR is also given at two different output levels, 4.2V and 10V. In each case the SNR corresponds to an output level just over 3uV, which is very good. I'm really pleased to see a 10V output option. The best processors like Storm, Trinnov and Lyngdorf have high output capability, maximising DNR, and potentially allowing direct connection to high performance / low gain amplifiers like Benchmark, Hypex, Purifi etc.

All these figures aren't QUITE as good as the best stereo DACs, but I think they're credible and more importantly, GOOD ENOUGH.

I tabled all the best multi-channel audio interfaces that could partner with the DPR-16 to perform DA conversion.
Many of them even have parametric EQ, though the Hyperion is almost certainly better.
If the APR-16 specs are accurate, it's comparable with the best of them, like RME, ExaSound, Antelope, Lynx, MOTU, PrismSound, DirectOut, Avid, and even Merging.
Therefore as long as you don't want to use HLC or HLP or AOIP, then the APR-16 is probably good enough IMHO.

 

[PGAMiami]

Well this was interesting. Hyperion originally claimed 130dB DNR / 120dB THDN for the APR-16, and I was sceptical about that.
They have clarified things, and now say this about the DACs, which appears to refer to the performance of the DAC chips themselves.



For the spec of the APR-16 unit itself, they now say:

View attachment 498386

0.0002% THD+N is -114dB. That's less than the chip spec, but better than any other AVP. Ever.

They also quote different SNR at different output voltage. First, because the DACs are most likely being used in current output mode (that's how high performance DACs perform best), so the DACs will feed I to V converters, and only after that can the performance be expressed relative to an output voltage. So that's a pretty good indication in itself that those aren't chip specs.

Second, the SNR is also given at two different output levels, 4.2V and 10V. In each case the SNR corresponds to an output level just over 3uV, which is very good. I'm really pleased to see a 10V output option. The best processors like Storm, Trinnov and Lyngdorf have high output capability, maximising DNR, and potentially allowing direct connection to high performance / low gain amplifiers like Benchmark, Hypex, Purifi etc.

All these figures aren't QUITE as good as the best stereo DACs, but I think they're credible and more importantly, GOOD ENOUGH.

I compiled a table of all the best multi-channel audio interfaces (ADCs & DACs) that I could find. These are the sort of products that could be partnered with the DPR-16 to perform DA conversion. Some of them even have parametric EQ, though the Hyperion is almost certainly better. The point of this is that the APR-16 specs, if accurate, are right up there with the better audio interfaces, like RME, ExaSound, Antelope, Lynx, MOTU, PrismSound, DirectOut, Avid, and even Merging.

Therefore if you accept that the Hyperion DSP is powerful to do what you want without using HLC / HLP, and you don't want to drive your speakers over AOIP, then the APR-16 is probably good enough in every respect and without qualification IMHO.

View attachment 498393

Very helpful. Thank you. Remaining question for me is how Dirac will be implementes. Is it like a basic, D&M implementation or instead more like Storm. I believe we’ve determined above that the DSPs in the APR-16 have about 1/2 the processing power of the 32 channel Storm, and therefore it should be enough for the 16 channel APR-16. Less taps than Trinnov, but Dirac ART can be a better fit for certain rooms like mine where multiple subs on the read wall are not practical. Other question I have is how to use Roon on the APR-16. HDMI works for multichannel Roon on the Denon 4800, but it’s very quirky, switches input whenever the Denon is turned on, results in audible noise on speakers even when this input is not selected if the turn on sequence doesn’t have the TV first. Strange. The Roon app built int9 Denon is not multichannel at this time.

 

[Highfidelity]

Blows my mind that they couldn’t achieve the performance of the dead simple circuit on the es9039Q2M eval board. I guess they should’ve just copied the BOM of the eval board. But engineers hate doing that because it’s too easy.

 

[Highfidelity]

There is no single x86 CPU from which one could generalize performance. We dont know how many taps the CPUs in the Trinnov can handle in the end when you really want to push settings. We only know that their CPUs are much more powerful also for DSP workloads than those FPGA chips fron analog devices that Storm et al use.
Why do they have that much power? Because some middle class Intel CPU is actually not that expensive and you want the units to be future proof. Most of the engineering work is to code for DSP work and decoders and build the custom Linux platform. Once you have that you are much more flexible than with FPGA chips and also because you dont have to use hard wired decoders. That is also why Storm uses the ADEC board now. This is not a specific decoder board, it uses an ARM CPU for which the decoder software can be programmed. With that respect they leaned a bit to the Trinnov approach.

This is the Storm ADEC board:

Networked Audio Processor 1 (NAP1)｜Networked Audio Processing｜Audio Solutions｜Elytone

NAP1 is a breakthrough in miniaturization and audio performance: a comprehensive audio system driven by a powerful multi-core application processor with on board memory, and power circuitry - all in a module about the size of a business card. NAP1 is designed to integrate networking, processing...

www.elytone.com

 

[Littletycoon]

Is the zman (ravenna) board in the dpr16?

 

[Highfidelity]

No they’re using Dante. So they’re either using a Brooklyn 3 or rolled their own Zynq based implementation.