diff options
Diffstat (limited to 'decoders/uart')
-rw-r--r-- | decoders/uart/pd.py | 25 |
1 files changed, 14 insertions, 11 deletions
diff --git a/decoders/uart/pd.py b/decoders/uart/pd.py index d59c6c3..7995e99 100644 --- a/decoders/uart/pd.py +++ b/decoders/uart/pd.py @@ -19,6 +19,7 @@ ## import sigrokdecode as srd +from math import floor, ceil ''' OUTPUT_PYTHON format: @@ -139,24 +140,24 @@ class Decoder(srd.Decoder): ) def putx(self, rxtx, data): - s, halfbit = self.startsample[rxtx], int(self.bit_width / 2) - self.put(s - halfbit, self.samplenum + halfbit, self.out_ann, data) + s, halfbit = self.startsample[rxtx], self.bit_width / 2.0 + self.put(s - floor(halfbit), self.samplenum + ceil(halfbit), self.out_ann, data) def putpx(self, rxtx, data): - s, halfbit = self.startsample[rxtx], int(self.bit_width / 2) - self.put(s - halfbit, self.samplenum + halfbit, self.out_python, data) + s, halfbit = self.startsample[rxtx], self.bit_width / 2.0 + self.put(s - floor(halfbit), self.samplenum + ceil(halfbit), self.out_python, data) def putg(self, data): - s, halfbit = self.samplenum, int(self.bit_width / 2) - self.put(s - halfbit, s + halfbit, self.out_ann, data) + s, halfbit = self.samplenum, self.bit_width / 2.0 + self.put(s - floor(halfbit), s + ceil(halfbit), self.out_ann, data) def putp(self, data): - s, halfbit = self.samplenum, int(self.bit_width / 2) - self.put(s - halfbit, s + halfbit, self.out_python, data) + s, halfbit = self.samplenum, self.bit_width / 2.0 + self.put(s - floor(halfbit), s + ceil(halfbit), self.out_python, data) def putbin(self, rxtx, data): - s, halfbit = self.startsample[rxtx], int(self.bit_width / 2) - self.put(s - halfbit, self.samplenum + halfbit, self.out_bin, data) + s, halfbit = self.startsample[rxtx], self.bit_width / 2.0 + self.put(s - floor(halfbit), self.samplenum + ceil(halfbit), self.out_bin, data) def __init__(self, **kwargs): self.samplerate = None @@ -189,7 +190,9 @@ class Decoder(srd.Decoder): # bitpos is the samplenumber which is in the middle of the # specified UART bit (0 = start bit, 1..x = data, x+1 = parity bit # (if used) or the first stop bit, and so on). - bitpos = self.frame_start[rxtx] + (self.bit_width / 2.0) + # The samples within bit are 0, 1, ..., (bit_width - 1), therefore + # index of the middle sample within bit window is (bit_width - 1) / 2. + bitpos = self.frame_start[rxtx] + (self.bit_width - 1) / 2.0 bitpos += bitnum * self.bit_width if self.samplenum >= bitpos: return True |