Mercurial > pygar

     1 #LyX 1.6.4 created this file. For more info see http://www.lyx.org/

     2 \lyxformat 345

     3 \begin_document

     4 \begin_header

     5 \textclass article

     6 \use_default_options true

     7 \language english

     8 \inputencoding auto

     9 \font_roman default

    10 \font_sans default

    11 \font_typewriter default

    12 \font_default_family default

    13 \font_sc false

    14 \font_osf false

    15 \font_sf_scale 100

    16 \font_tt_scale 100

    17 

    18 \graphics default

    19 \paperfontsize default

    20 \use_hyperref false

    21 \papersize default

    22 \use_geometry false

    23 \use_amsmath 1

    24 \use_esint 1

    25 \cite_engine basic

    26 \use_bibtopic false

    27 \paperorientation portrait

    28 \secnumdepth 3

    29 \tocdepth 3

    30 \paragraph_separation indent

    31 \defskip medskip

    32 \quotes_language english

    33 \papercolumns 1

    34 \papersides 1

    35 \paperpagestyle default

    36 \tracking_changes false

    37 \output_changes false

    38 \author "" 

    39 \end_header

    40 

    41 \begin_body

    42 

    43 \begin_layout Title

    44 Multi-Voice Audio Playback

    45 \end_layout

    46 

    47 \begin_layout Author

    48 Laurel Pardue, Robert McIntyre

    49 \end_layout

    50 

    51 \begin_layout Section*

    52 Micro-Architecture

    53 \end_layout

    54 

    55 \begin_layout Subsection*

    56 A quick summary of our system

    57 \end_layout

    58 

    59 \begin_layout Standard

    60 The audio data (

    61 \begin_inset Quotes eld

    62 \end_inset

    63 

    64 samples

    65 \begin_inset Quotes erd

    66 \end_inset

    67 

    68 ) start in the memory, but are soon pulled into action by the DMA (direct

    69  memory access).

    70  The DMA sends the samples to a chain of 0 or more 

    71 \emph on

    72 soft-cores

    73 \emph default

    74 , where they are transformed according to the soft-cores' algorithms.

    75  After running the gauntlet of soft-cores, the samples flow first to a buffering

    76  FIFO, and finally to a 

    77 \emph on

    78 mixer

    79 \emph default

    80 , which sends the samples off to be played by speakers of stored in a file.

    81 \end_layout

    82 

    83 \begin_layout Subsection*

    84 A more detailed look a our system

    85 \end_layout

    86 

    87 \begin_layout Standard

    88 Our audio pipeline starts its life with whatever music we want to play already

    89  in memory.

    90  The music has multiple tracks, or 

    91 \begin_inset Quotes eld

    92 \end_inset

    93 

    94 voices,

    95 \begin_inset Quotes erd

    96 \end_inset

    97 

    98  which we arrange on the memory in contiguous sequences.

    99  Therefore, each voice starts at its own start-address, and continues on

   100  for as long as it needs.

   101  We cap each voice with an end-of-file marker.

   102  We make every voice the same length to simplify timing issues, even though

   103  that might mean that there is a lot of wasted space in the form of dead

   104  areas with no sound.

   105 \end_layout

   106 

   107 \begin_layout Standard

   108 When the processing starts, the DMA reads in each voice and sends the audio

   109  data (

   110 \begin_inset Quotes eld

   111 \end_inset

   112 

   113 samples

   114 \begin_inset Quotes erd

   115 \end_inset

   116 

   117 ) off to a chain of soft-cores.

   118  The DMA knows to which chain the samples should be forwarded by looking

   119  up the samples' voice in a hash table read at startup.

   120  Once the samples enter the soft-core chain, the first soft-core in line

   121  takes the samples and performs some sort of audio operation, producing

   122  modified audio samples.

   123  These samples feed into the next soft-core which performs a different algorithm.

   124  As condition for functional correctness, we require the algorithm that

   125  each soft-core contains to be able to produce audio samples at the desired

   126  44khz rate, but we allow an arbitrary setup time for the soft-core to get

   127  ready.

   128  That way, a pipelined processor can take its time and fill up its pipeline

   129  completely with no worries.

   130  Once the soft-core starts producing samples, it must produce them at a

   131  rate of at least 44khz forever.

   132  If every soft-core in the chain obeys this timing condition, then the whole

   133  chain will as well, so the entire chain of soft-cores can be treated as

   134  a single big soft-core for timing purposes.

   135  Our soft-cores are just Lab 5 processors preloaded with specific audio

   136  processing algorithms.

   137 \end_layout

   138 

   139 \begin_layout Standard

   140 Eventually, the samples make their way to the end of the chain of soft-cores.

   141  Every chain of soft-cores ends in a FIFO.

   142  For this project, we will have 12 separate voices.

   143  Therefore, there will be 12 soft-core chains, each with a FIFO at its end,

   144  one for each voice.

   145  Each of the 12 FIFOs leads to the mixer, which only begins operation once

   146  all 12 FIFOs are full.

   147  Because of the timing constraints we impose on our soft-cores, once every

   148  FIFO has at least one sample, all 12 will forever more be able to supply

   149  samples at 44khz.

   150  The mixer therefore waits until all 12 FIFOs have elements and then reads

   151  samples from all FIFOS at a rate of 44khz.

   152  The mixer combines the samples through simple addition and scaling, creating

   153  a single stream of sample data, which is sent back from the FPGA to the

   154  host computer where we play the stream and/or save it to a file.

   155  The mixer ensures that it outputs samples at a rate of 44khz by running

   156  everything through its own internal buffer FIFO clocked at 44khz.

   157 \end_layout

   158 

   159 \end_body

   160 \end_document