function | V/P | Shift
Bits |
Bits | Sat | notes |

A/D sample | V | 12 | Adc
Ovf |
||

PolyFilter/ffts | V | 18 | 19 bits arithemetic. | ||

Pshift |
V | Down | 18 | fft
Ovf |
13 bits mask. Each bitdown shifts after a butter fly stage.
B0 last butterfly(size2). B12 first butter fly stage. |

Shift
Vshift |
V | 3Up | 32 | Vsh
Sat |
After fft before stokes upshift values. |

Compute pwr | P | 32 | |||

Dshift |
P | 4Dwn | 32 | down shift prior to accumulation. | |

Accumulation | P | 40 | Acc
Sat |
accumulate in 40 bit numbers. Assume 32 bit input placed in lower 32 bits of accumulator | |

Ashift |
P | 3Up | 32 | Ash
Sat |
uppper most 32 bits of accumulator are up shifted in 32 bit word. This is needed for packing 8 or 16 bits. |

- The input A/D should have an rms that is probably 5 counts or more to be above the digitization noise. You should allow a few extra bits so you can setup the power levels at high za and then continue observing at low za (where Tsys is lower).
- The output spectrum should have enough counts so that the noise in a channel covers a few counts. This is determined by the radiometer equation: dT/T=1./Sqrt(bw*time). If T is the number of counts in the spectral channel then Dt=SpectralChanVal/sqrt(bw*time).
- For a 1 second integration, 8K channels, and 170 Mhz :
- 1./sqrt(bw*time)=.0069.
- 1/.0069= 144.
- So you will need a spectral channel value of 144 to have 1 count on the noise.
- 5 counts would be about 720.
- This holds for the edges of the spectrum where the filter shape is falling off.

Testing the different setting for PFB saturation/underflow.

A/D
Lvls |
Pshift | pfbOvr
L,H |
VSh | Dsh | Ash | Spc
Med |
Notes |

11,10 | 0x1800
0x1a00 |
0,3
0,0 |
0 | 0 | 0 | need >=3 shifts | |

26,22 | 0x1300
0x0013 0x1600 |
0,0
12,7 0,0 |
0 | 0 | 0 | need >=3 shifts | |

47,40 | 0x1600
0x1700 |
7,8
0,0 |
0 | 0 | 0 | need >=4 shifts |