The importance of processing power in Home Theater

An architectural choice: CPU vs. DSP

The raw computational power of any processor is measured in giga floating point operations per second (GFLOPs). (One GFLOP equals one billion floating point operations per second). Floating point operations are used for precise mathematical computations needed for complex operations.

Before we examine how Trinnov uses this raw power, we will explore some of the significant differences between chip-based and CPU-based processing power. Most AV processors use DSP chips with less than 3 GFLOPs, while the most powerful can reach almost 20 GFLOPs. 

By comparison, the CPU currently used in the Altitude32 delivers 240 GFLOPs and as many as 422 GFLOPs using Intel’s turbo boost technology. The Altitude16, having fewer channels, still retains about 211 GFLOPs (275 GFLOPs with turbo boost). For comparison purposes, we will use the lower numbers here (which are still more than ten times the computational power of the most potent competitive system). 

Any surround processor must perform a lot of complex math in real-time. As crucial as raw computational power is, however, GFLOPs do not tell the whole story.

Processing power can only be used if the processor is not waiting for data and program instructions to execute. Unfortunately, external memory is too slow to be fetched directly by the processor, whether it is a CPU or a DSP chip. 

This problem is resolved using internal or “cache” memory directly on the processor chip. This high-speed storage layer temporarily holds frequently accessed data and instructions. Cache memory is crucial in real-time audio processing to ensure smooth and uninterrupted signal processing. If there is insufficient internal memory, the processor will waste many processing cycles waiting for the missing data to be fetched from external memory. Waiting for critical information wastes the system's available processing power.

For example, high-resolution FIR filters are extremely memory-intensive. However, they are also uniquely powerful tools for correcting both frequency response and timing errors in audio and are essential for effective room correction. We’ll get to the audible results of such filters in the upcoming section. Still, you can see in the chart below that the available cache memory of the Altitude’s advanced CPUs dwarfs anything found in even the most potent DSP chips.

Finally, there are significant advantages to doing all the required processing in one place. The Altitude’s single-processor architecture shares the internal memory among all of the internal processing cores, so every piece of data is instantly available to all the cores, eliminating any latency and processing overhead.

By contrast, many surround processors must use multiple DSP chips in arrays of two or four chips. While this is significantly less costly than having the power all in one CPU, it introduces several significant limitations. As you might imagine, the interconnections between chips doing disparate are complex and introduce significant synchronization issues between chips. This, in turn, often creates latency and processing overhead that affects the array's real-time processing performance.

Moreover, writing and optimizing software for parallel processing systems is more complex. Breaking down an advanced audio algorithm into independent pieces that run on separate processors is not straightforward.

For all these reasons, a single processor with the same theoretical processing power outperforms a processor array in most real-world applications.

Audio performance and applications

Ideally, high-performance processors must simultaneously support high-resolution audio (natively processing signals of 96k and even 192k), high channel count rendering and upmixing (up to 34 unique channels), and advanced Room Correction, including Active Acoustics.

Supporting all of these requirements simultaneously requires exceptionally high processing power. Using CPU-based architectures rather than state-of-the-art DSP engines reduces compromises. 

These compromises in chip-based DSP designs include: 

• Limiting the number of rendered immersive sound channels, limiting spatial resolution.
• Limiting the FIR filter length to 512 or 3000 taps (which, in turn, limits the FIR frequency resolution to 94 Hz or 16 Hz, respectively) offers insufficient resolution at low frequencies. By comparison, the Altitude runs acoustic optimization filters up to 32768 taps at 48 kHz while processing all channels. This provides 11 times more frequency resolution (1.46Hz) and allows control in the time domain over an acoustic path 11 times longer (680ms instead of 62ms).
• Giving up processing high-resolution audio at the native 96kHz or 192kHz audio processing frequency. All available DSP-based platforms are forced to downsample any incoming audio to 48kHz, reducing available bandwidth (which obviates the benefits of high-res audio files cherished by audiophiles). Only Altitude processors have the required processing resources to achieve high-resolution audio processing at the native audio frequency, with FIR filters up to 131,072 taps at 192kHz. 
• Settling for single-precision 32-bit floating point operations rather than the no-compromise 64-bit floating point operations required to offer the numerical precision that guarantees artifact-free audio processing. Double-precision processing requires double the processing power.
• Excluding some loudspeakers from the Digital Room Correction or Active Acoustics processing.

If each of these processing-saving tricks has a small audible impact when taken individually, they significantly impact the overall experience when combined. Only the Altitude processors have the processing resources to avoid all these compromises and ensure a fully optimized signal path. 

Processing headroom for the future

With so many processing tricks required to work with today’s technologies, chip-based DSP designs have little headroom for new technologies. After all, these new technologies are likely to require even more processing power, as they have consistently done for the past few decades.

The benefit of the Altitude platform and its CPU-based architecture is obvious: it avoids the necessity of these Memory/GFLOPs-saving tricks (and associated compromises). It preserves significant processing headroom to handle future, more demanding processing algorithms. 

This exceptional sustainability is otherwise unknown in the world of surround sound processors. Non-Trinnov processors are effectively “consumables” in multichannel systems since they must be replaced every few years to keep up with technological growth and innovation.

Thanks to the Altitude platform's unparalleled power and unique architecture, most new technologies have been incorporated via simple software updates rather than product replacement.

CEDIA recognized the longevity and long-term value of the Altitude platform in its prestigious Hall of Fame Award, which was given to Altitude32 in 2023.

So, why is Trinnov alone in using CPU-based processing?

With all of the advantages of Trinnov’s approach to designing advanced audio processors, one wonders why more companies have not followed this path.

Trinnov’s processors are radically unlike a “normal” PC running audio software. They derive from the research platform the company's founders developed for doing the advanced research into how to capture and reproduce complex, three-dimensional sound fields for which they created the company. Far from a “Home Theater PC,” Trinnov’s processors rely on three advanced methods refined over more than twenty years of development:

• Trinnov-OS is a unique operating system. The Altitude processor relies on a bespoke quasi-real-time operating system with a specifically-tuned, ultra-reactive scheduler that allows extremely fast context switching.
• Kernel level processing. Critical aspects of the audio processing chain are implemented at the kernel level, which grants them ultimate priority, the same as the most fundamental functions of the operating system. Other processes running on the same CPU do not disturb this audio processing.
• Multithreading. The audio processing software is designed for parallel processing, taking advantage of modern CPUs' multi-core capabilities.

Other companies have not invested years developing such state-of-the-art technologies, instead relying on assembling products from readily available, off-the-shelf parts. This approach might be easier and less costly, but it forces compromises that Trinnov Audio's founders will not accept.

In stark contrast, Trinnov engineers write the audio software entirely in-house, including the advanced Trinnov OS (which is based on Linux but has evolved into a custom, real-time audio processing powerhouse). This investment allows us to preserve complete control over the processing and guarantees end-to-end audio quality and efficiency.

Summary

We all recognize that the Altitude platform requires a higher initial investment than most competitors. However, this is a “buy once, cry once” investment in lasting value rather than spending money every few years on a home cinema consumable. 

This initial investment provides lasting value through regular (and free) software updates, ensuring ample processing headroom for new technologies that will undoubtedly emerge over time. It also enables sophisticated capabilities that multi-DSP designs cannot match while giving Trinnov complete control over its signal processing and path.

Rather than relying on DSP engineers at mass-market chip companies to determine what features are “good enough,” Trinnov’s engineers maintain complete control over the product's capabilities.  Even when new hardware is necessary, such as with the evolution of the HDMI standard, the Altitude’s modular design makes it easy and relatively inexpensive to replace the board instead of the entire processor. 

The Altitude platform offers unique value among otherwise disposable surround processors.