PMC-SpaceWire Dynamic Engineering 4-Channel SpaceWire Interface Board

The Dynamic Engineering PMC-SpaceWire 4-Channel SpaceWire Interface Board implements SpaceWire in a convenient PMC format. Four fully independent and highly programmable LVDS IO ports are provided by the PMC-SpaceWire design. The SpaceWire protocol is advanced with link testing, error handling, command and data protocols built in. The SpaceWire electrical interface is point-to-point with token based flow control. With the SpaceWire protocol it is easy to build a hierarchical architecture system with routers or a home-run wired system. The user's equipment can interact with any other node in the system. PMC-SpaceWire provides a bridge from PCI to SpaceWire. Port based DMA offloads the CPU and increases performance in the user's system. Time Code support provides a complete solution.

Each port has memory to support RX and TX functions. BK models have more than 32 MBytes per port. The memory is organized as FIFOs 32 bits wide to optimize data transfer from the PCI/PCIe bus. The FIFOs are a combination of internal block RAM and attached DDR memory.

Multiple programmable interrupts are avaialble. The FIFO flags are supported for interrupt driven or polled operation. Programmable Almost Full, Programmable Almost Empty, plus Full, Half Full, and Empty. The programmable flags can be set to any debth to allow software a programmable delay from event to empty or full. The interrupts can be very useful when sending or receiving messages larger than the FIFO size. In addition interrupts and status are available for packet completion, various error conditions etc.

The FIFO design supports loop-back and can be used for Built In Test [BIT]. Memory can be accessed with target or DMA accesses. The host side interface is optimized with independent DMA controllers for each port (eight total).

SpaceWire features automatic link testing when a node is brought on line - released for operation via software. The transmitter sends NULLs until NULLs are received and then sends FCTs until FCTs are received. The initial frequency is specified [ECSS-E-ST-50-12C] to be 10 MHz. Once the auto link test has passed, the hardware can switch to the programmed rate. A local oscillator is used to supply the 10 MHz reference clock for initialization. A programmable PLL is used to support operation once the link is established. The PLL is programmed via software and is referenced to the oscillator.

The SpaceWire protocol has flow control. The local memory on PCIe-SpaceWire will not overrun. The buffering issue may be upstream - if there is an unbuffered or under-buffered data source that will loose data if the link is throttled back the large storage capability will help. The larger FIFOs will provide more room to accumulate data if the system is not ready to allow PCIe-SpaceWire to DMA transfer immediately. The DMA transfers are higher bandwidth than the SpaceWire transfers allowing the system to catch-up once the DMA transfers restart. At 100 MHz and 80% data transfer the 8Mx32 FIFO will provide about 3.2 seconds of data storage per port. The host would have to ignore the PCIe-SpaceWire interrupt requests for longer than 3.2 seconds before throttling would occur. In most systems this is sufficient time to stay out of the throttled mode. To unload a full FIFO with DMA running it will take 8M/50M = .16mS => the unload to load multiplier is strongly in favor of the host side allowing the host to catch-up to the SpaceWire link once the delaying event has passed. The actual DMA time will be a bit higher due to system interaction to initialize.