In describing an ideal audio workstation, users want maximum power and speed from a system that’s forward- and backward-compatible, versatile, expandable and safe from becoming obsolete in the near future. And while we’re at it, let’s throw in greatly reduced latency and lower cost.
With the unveiling of its new Dream II series, Fairlight (www.fairlightau.com) may have come closer to the ideal DAW. We’ve heard that one before, but this one’s different — very different. Based on the company’s new CC-1 (Crystal Core technology) card, the Dream II Series uses a large-scale Field-Programmable Gate Array (FPGA) rather than the traditional ASIC- or CPU/DSP-based approach. The CC-1 architecture is implemented on a new-generation FPGA chip manufactured by Altera Corporation (www.altera.com).
“In early 2004, we were looking into new directions in engine technology,” says Fairlight’s CTO, Tino Fibaek. “Rather than just going brute-force with bigger chips, we looked for an alternative, which we found in a specific family of FPGAs. CC-1 has a smallish PCB with a single, large FPGA chip, but we can change the contents of that chip on the fly. You might have one layout of the chip on one version of the software, but in the next software revision have a completely different layout of the chip. The equivalent of that today is removing the entire PCB from your old engine and installing a completely new one.
Fairlight’s CC-1 card with CMI expansion board provides hundreds of I/Os from eight MADI ports.
“From that point of view, our system has a lot of future-proofing built in,” Fibaek continues. “The idea was to develop a system so existing Fairlight customers could easily upgrade without having to rework their rooms or spend a lot of time retraining their staff. The change is primarily on the audio engine and I/O side — the user interface is unchanged and the surface controller technology doesn’t change at all, and owners can upgrade to the new technology quite easily. And as users had fairly large libraries of media assets and projects, we wanted to provide access to all of them. Dream II is upward and downward file-compatible with the old systems.”
INSIDE THE CRYSTAL CORE
The Crystal Core platform is named for the crystalline-styled interconnect scheme used at every level of system architecture. Processing blocks connect to other blocks, forming a lattice that scales the project’s processing/routing demands to different nodes for efficient power distribution. As an example, a simple node handling I/O duties requires less resources than a complex node doing multichannel equalization, and the system usage is optimized by allocating the right amount of resources to each job. In traditional systems, the architecture is locked into standard configurations, with resources hardwired to signals that may not need them.
Another unique aspect of Crystal Core is its ability to combine floating point and fixed signal processing computations, using the advantages of both approaches. For example, 72-bit fixed point can be used for the equalization node, while 36-bit floating point is used in the mixing node. In other situations, such as audio metering, this level of precision is not essential and tasks could be handled by fixed point 16-bit processing. Fairlight refers to this as Dynamic Resolution Optimization (DRO) architecture and has patented the process.
SX-48 modular remote I/O accommodates a MADI I/O card and up to six 8-channel audio cards.
Described as a “media-optimized FPGA architecture” forming an aggregation of IP cores, CC-1 is a PCI Express card fitted inside a host PC. This single card can provide 200-plus channels of audio recording/editing/mixing into 72 mix buses, with each channel equipped with 12 aux sends, eight bands of EQ and three stages of dynamics (limiter, compressor and expander), along with uncompressed HD video. If that’s not enough horsepower, four CC-1 cards offer 960 channels into 320 buses, with no theoretical limit to the number of linkable processors, although Fairlight is initially only supporting one card per system. But the cost savings are significant. The CC-1’s single FPGA yields the same performance as Fairlight’s Constellation system, which uses 64 Analog Devices SHARC DSP chips.
In larger facilities, Fairlight’s new Media Highway protocol incorporated into the system can transmit multiple data streams in different formats, bit resolutions and sampling frequencies simultaneously, with clock transmission and recovery to/from any point on a network. The Media Highway protocol connects using a facility’s existing cabling.
The system packs a lot of I/O on a single card. The CC-1’s directly accessible I/O connections include BNC MADI input and output ports (64 channels each) and a link to a small, single-rackspace SX-20 I/O breakout box. Designed to provide all of the I/O for a smaller system (or for convenient connections in a control room in a larger system), the SX-20 has two mic/line inputs, two line inputs, 12 analog outputs, four S/PDIF ins, eight S/PDIF outs, dual 9-pin machine control, video sync and LTC timecode. Another connector links to a small expansion board (Model CMI) that adds another three MADI inputs and outputs for a total of four MADI I/Os, plus the connections on the breakout box.
In keeping with Crystal Core’s chameleon-like capabilities to change functionality, the MADI ports are actually multiplexed data connectors and can become either MADI for audio handling or SDI video data streams. The MADI ports can be connected to any third-party MADI device or to Fairlight’s SX-48 modular remote I/O.
Housed in a two-rackspace frame, the SX-48 accommodates one MADI I/O card and up to six 8-channel audio cards, providing AES EBU I/O with sample rate conversion, eight analog inputs or eight analog outputs. Up to four SX-48s can be connected for up to 192 physical I/Os, with the system entirely configurable to the user’s needs.
AND INTO THE FUTURE
“The world has changed and we can’t keep looking at products using an old approach,” Fibaek explains. “We can talk about the number of channels, increased speed, et cetera — and that’s not a bad start — but most of the benefits of FPGA technology stem from its concept of virtual hardware. Every bit of design we’re doing is making sure that it’s scalable toward new technologies. FPGA is a fast-moving market at the moment and what we have today is perfect for what we’re doing, but we have to be ready to jump on the next generation.
“Not only could the features of an application change — say, instead of having three stages of dynamics now, we could have four in the future — but the fundamental operation of the board could change completely. Instead of being very heavily focused on audio, it could be more middle-of-the-road on audio, with more video processing.” This opens entirely new options to users, especially those in multitasking facilities. For example, a system could be taken down and completely reprogrammed to change from a large-scale audio mixer into something like a video color-correction device in a matter of seconds.
The process can also take place in real time. An FPGA’s ability to upload partial algorithms while in session could allow continuous reconfiguration while operating. This would be ideal in the context of a user-controlled system, where tasks can change during a session.
“Changing the content of a FPGA is the equivalent of changing and re-routing an entire PCB,” says Fibaek. “And along with changes in the front-end software and the user interface, we can continue to effectively develop the hardware, even though we shipped it to our customers a long time ago. We can continue to give users more features over time, which to me is the most exciting thing about this technology.”
Fairlight is also exploring the expansion of the platform by opening the hardware and software architecture to other manufacturers and third-party (hardware and software) developers. The Crystal Core — powered product family will debut at this month’s AES (Booth #326) in the form of a single CC-1 card that connects with any of Fairlight’s Satellite, Station, Constellation and Anthem surface technologies. Deliveries are expected to begin later this year.
Inside Fairlight’s Crystal Core Technology
Anthem Dream is one of the Fairlight systems that benefits from the company’s new Crystal Core technology.
After decades of digital products based on DSP approaches, Fairlight’s new CC-1 (Crystal Core) technology takes a new direction that processes data in a massive Field Programmable Gate Array (FPGA) architecture that essentially functions as a purpose-built media processing chip with low latency performance. The power of this single-FPGA system will support more than 200 audio channels with HD video, and larger systems can be created over a wide, fast data highway to interconnect across chips, between computers, or from room to room.
The “crystal” part of the name comes from the crystalline interconnect scheme used throughout the architecture. Processing blocks connect in three dimensions to other blocks, forming an extensible lattice that scales as required to meet the volume processing demands of media applications. The processing blocks themselves are FPGAs, each programmed with the embedded intelligence required to fulfill the demands of the particular product or system being addressed. Fairlight’s pending patent defines an architectural arrangement of IP cores supporting real-time audio and video processing systems.
The Crystal architecture is based on the use of high-density FPGAs. Each of these chips contain millions of simple logic parts, that can either statically or dynamically upload a program that configures those parts into complex arrangements for high-speed computer applications. These applications will be used to replace media signal processing hardware that was previously composed of dedicated components or partially-programmable digital hardware.
FPGAs can be programmed to form a great variety of standard logic components and I/O ports that can be built into complex sub-systems. Larger programs can even embed whole processors exactly emulating commonly-used DSPs or CPUs into a part of the chip while other parts function differently. This makes it possible, with the addition of a few extra components—such as RAM and some standard computer bus interface chips—to build a complete system using one programmable device.
Each processor chip is programmed to form a hub-node structure, where a single hub connects a number of nodes. The hub and each of the nodes are individual IP cores, programmed in hardware to perform a specific task within a media system. A hub is a signal routing core, effectively a broadcast source transmitting data to all receptor nodes and receiving data back from them. A typical large audio mixing system may use hubs broadcasting 2,000 audio signals of 36-bit depth at 48,000 samples per second, or its data equivalent in other media or at other sample rates. And larger hubs can be programmed as needed.