10

1772E–DRMSD–10-Apr-06

ATSAM9708

This data bus cannot drive the PC bus directly. External buffers and an external decoder

(PAL) or plug and play IC are required to map the 16-bit I/O addresses and AEN from the PC

into the three address lines and chip select from the ATSAM9708.

The PDSP#1 responds on addresses 0 to 3 (A2A1A0 = 0XX), while PDSP#2 responds on

addresses 4 to 7 (A2A1A0 = 1XX).

Each PDSP parallel interface supports a byte-wide I/O interface and a 16-bit port dedicated to

burst transfers.

The byte-wide I/O interface is normally used to implement a MPU-401 UART-mode compati-

ble interface. It is specified by address A1A0 = 0X, address 00 being the data register,

address 01 being the status/control registers. Besides the standard two status bits of the

MPU-401, two additional bits are provided to expand the MPU-401 protocol.

Address A1A0 = 10 specifies a 16-bit I/O port. It is mainly used for burst audio transfers

to/from the PC using very efficient PC instructions like REP OUTSW or REP INSW which

operate at maximum ISA bus bandwidth. This port may also be used for fast program or sound

bank uploads.

10. DSP Cache RAM

The memory management unit (MMU) allows external ROM and/or RAM memory resources to

be shared between the two DSPs and the two P16 control processors. This allows a single

device (i.e., DRAM) to serve as sample memory storage/delay lines for the DSPs and as pro-

gram storage/data memory for the P16 control processors.

The DSP cache RAM allows a dramatic reduction in the traffic with the external ROM/RAM,

allowing use of standard 90 ns ROM parts with sampling frequencies up to 48 kHz. Average

access request rate to external memory is only one for every two frames for each slot, which

gives 64 accesses per synthesis frame. The MMU can provide up to 169 memory accesses

per frame, which leaves over 100 accesses free per frame to be used by the P16 processors.

This means that under full 128-voice polyphony traffic conditions, each P16 instruction aver-

age execution time is around 400 ns at 48 kHz sampling frequency.

128-voice polyphony can be assured only when all samples are played at nominal frequency

or down-transposed. Simultaneously playing a large number of up-transposed samples can

adversely affect polyphony. For more details of possible polyphony for a given application,

please refer to the application note “ATSAM9708 Memory Management Unit”.