1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
|
#include <unistd.h>
#include <stdbool.h>
#include <math.h>
#include <stdlib.h>
#include <assert.h>
#include <arm_math.h>
#include "freq_meas.h"
#include "sr_global.h"
#include "dsss_demod.h"
#include "simulation.h"
#include "generated/dsss_gold_code.h"
#include "generated/dsss_butter_filter.h"
/* Generated CWT wavelet LUT */
extern const float * const dsss_cwt_wavelet_table;
struct iir_biquad cwt_filter_bq[DSSS_FILTER_CLEN] = {DSSS_FILTER_COEFF};
void debug_print_vector(const char *name, size_t len, const float *data, size_t stride, bool index, bool debug);
static float gold_correlate_step(const size_t ncode, const float a[DSSS_CORRELATION_LENGTH], size_t offx, bool debug);
static float cwt_convolve_step(const float v[DSSS_WAVELET_LUT_SIZE], size_t offx);
static float run_iir(const float x, const int order, const struct iir_biquad q[order], struct iir_biquad_state st[order]);
static float run_biquad(float x, const struct iir_biquad *const q, struct iir_biquad_state *const restrict st);
static void matcher_init(struct matcher_state states[static DSSS_MATCHER_CACHE_SIZE]);
static void matcher_tick(struct matcher_state states[static DSSS_MATCHER_CACHE_SIZE],
uint64_t ts, int peak_ch, float peak_ampl);
static void group_received(struct dsss_demod_state *st);
#ifdef SIMULATION
void debug_print_vector(const char *name, size_t len, const float *data, size_t stride, bool index, bool debug) {
if (!debug)
return;
if (index) {
DEBUG_PRINTN(" %16s [", "");
for (size_t i=0; i<len; i++)
DEBUG_PRINTN("%8zd ", i);
DEBUG_PRINTN("]\n");
}
DEBUG_PRINTN(" %16s: [", name);
for (size_t i=0; i<len; i++)
DEBUG_PRINTN("%8.5f, ", data[i*stride]);
DEBUG_PRINTN("]\n");
}
#else
void debug_print_vector(const char *name, size_t len, const float *data, size_t stride, bool index, bool debug) {}
#endif
void dsss_demod_init(struct dsss_demod_state *st) {
memset(st, 0, sizeof(*st));
matcher_init(st->matcher_cache);
}
void dsss_demod_step(struct dsss_demod_state *st, float new_value, uint64_t ts) {
//const float hole_patching_threshold = 0.01 * DSSS_CORRELATION_LENGTH;
st->signal[st->signal_wpos] = new_value;
st->signal_wpos = (st->signal_wpos + 1) % ARRAY_LENGTH(st->signal);
/* use new, incremented wpos for gold_correlate_step as first element of old data in ring buffer */
for (size_t i=0; i<DSSS_GOLD_CODE_COUNT; i++)
st->correlation[i][st->correlation_wpos] = gold_correlate_step(i, st->signal, st->signal_wpos, false);
st->correlation_wpos = (st->correlation_wpos + 1) % ARRAY_LENGTH(st->correlation[0]);
float cwt[DSSS_GOLD_CODE_COUNT];
for (size_t i=0; i<DSSS_GOLD_CODE_COUNT; i++)
cwt[i] = cwt_convolve_step(st->correlation[i], st->correlation_wpos);
float avg = 0.0f;
for (size_t i=0; i<DSSS_GOLD_CODE_COUNT; i++)
avg += fabs(cwt[i]);
avg /= (float)DSSS_GOLD_CODE_COUNT;
/* FIXME fix this filter */
//avg = run_iir(avg, ARRAY_LENGTH(cwt_filter_bq), cwt_filter_bq, st->cwt_filter.st);
float max_val = st->group.max;
int max_ch = st->group.max_ch;
int max_ts = st->group.max_ts;
bool found = false;
for (size_t i=0; i<DSSS_GOLD_CODE_COUNT; i++) {
float val = cwt[i] / avg;
if (fabs(val) > fabs(max_val)) {
max_val = val;
max_ch = i;
max_ts = ts;
if (fabs(val) > DSSS_THESHOLD_FACTOR)
found = true;
}
}
/* FIXME: skipped sample handling here */
matcher_tick(st->matcher_cache, ts, max_ch, max_val);
if (found) {
/* Continue ongoing group */
st->group.len++;
st->group.max = max_val;
st->group.max_ch = max_ch;
st->group.max_ts = max_ts;
return;
}
if (st->group.len == 0)
/* We're between groups */
return;
/* A group ended. Process result. */
group_received(st);
/* reset grouping state */
st->group.len = 0;
st->group.max_ts = 0;
st->group.max_ch = 0;
st->group.max = 0.0f;
}
/* Map a sequence match to a data symbol. This maps the sequence's index number to the 2nd to n+2nd bit of the result,
* and maps the polarity of detection to the LSb. 5-bit example:
*
* [0, S, S, S, S, S, S, P] ; S ^= symbol index (0 - 2^n+1), P ^= symbol polarity
*
* Symbol polarity is preserved from transmitter to receiver. The symbol index is n+1 bit instead of n bit since we have
* 2^n+1 symbols to express, one too many for an n-bit index.
*/
uint8_t decode_peak(int peak_ch, float peak_ampl) {
return (peak_ch<<1) | (peak_ampl > 0);
}
void matcher_init(struct matcher_state states[static DSSS_MATCHER_CACHE_SIZE]) {
for (size_t i=0; i<DSSS_MATCHER_CACHE_SIZE; i++)
states[i].last_phase = -1; /* mark as inactive */
}
/* TODO make these constants configurable from Makefile */
const int group_phase_tolerance = (int)(DSSS_CORRELATION_LENGTH * 0.10);
void matcher_tick(struct matcher_state states[static DSSS_MATCHER_CACHE_SIZE], uint64_t ts, int peak_ch, float peak_ampl) {
/* TODO make these constants configurable from Makefile */
const float skip_sampling_depreciation = 0.2f; /* 0.0 -> no depreciation, 1.0 -> complete disregard */
const float score_depreciation = 0.1f; /* 0.0 -> no depreciation, 1.0 -> complete disregard */
const int current_phase = ts % DSSS_CORRELATION_LENGTH;
const int max_skips = TRANSMISSION_SYMBOLS/4*3;
for (size_t i=0; i<DSSS_MATCHER_CACHE_SIZE; i++) {
if (states[i].last_phase == -1)
continue; /* Inactive entry */
if (current_phase == states[i].last_phase) {
/* Skip sampling */
float score = fabs(peak_ampl) * (1.0f - skip_sampling_depreciation);
if (score > states[i].candidate_score) {
/* We win, update candidate */
assert(i < DSSS_MATCHER_CACHE_SIZE);
states[i].candidate_score = score;
states[i].candidate_phase = current_phase;
states[i].candidate_data = decode_peak(peak_ch, peak_ampl);
states[i].candidate_skips = 1;
}
}
/* Note of caution on group_phase_tolerance: Group detection has some latency since a group is only considered
* "detected" after signal levels have fallen back below the detection threshold. This means we only get to
* process a group a couple ticks after its peak. We have to make sure the window is still open at this point.
* This means we have to match against group_phase_tolerance should a little bit loosely.
*/
if (abs(states[i].last_phase - current_phase) == group_phase_tolerance + DSSS_DECIMATION) {
/* Process window results */
assert(i < DSSS_MATCHER_CACHE_SIZE);
assert(0 <= states[i].data_pos && states[i].data_pos < TRANSMISSION_SYMBOLS);
states[i].data[ states[i].data_pos ] = states[i].candidate_data;
states[i].data_pos = states[i].data_pos + 1;
states[i].last_score = score_depreciation * states[i].last_score +
(1.0f - score_depreciation) * states[i].candidate_score;
states[i].candidate_score = 0.0f;
states[i].last_skips += states[i].candidate_skips;
if (states[i].last_skips > max_skips) {
states[i].last_phase = -1; /* invalidate entry */
} else if (states[i].data_pos == TRANSMISSION_SYMBOLS) {
/* Frame received completely */
handle_dsss_received(states[i].data);
states[i].last_phase = -1; /* invalidate entry */
}
}
}
}
static float gaussian(float a, float b, float c, float x) {
float n = x-b;
return a*expf(-n*n / (2.0f* c*c));
}
static float score_group(const struct group *g, int phase_delta) {
/* TODO make these constants configurable from Makefile */
const float distance_func_phase_tolerance = 10.0f;
return fabsf(g->max) * gaussian(1.0f, 0.0f, distance_func_phase_tolerance, phase_delta);
}
void group_received(struct dsss_demod_state *st) {
const int group_phase = st->group.max_ts % DSSS_CORRELATION_LENGTH;
/* This is the score of a decoding starting at this group (with no context) */
float base_score = score_group(&st->group, 0);
float min_score = INFINITY;
ssize_t min_idx = -1;
ssize_t empty_idx = -1;
for (size_t i=0; i<DSSS_MATCHER_CACHE_SIZE; i++) {
/* Search for empty entries */
if (st->matcher_cache[i].last_phase == -1) {
empty_idx = i;
continue;
}
/* Search for entries with matching phase */
/* This is the score of this group given the cached decoding at [i] */
int phase_delta = st->matcher_cache[i].last_phase - group_phase;
if (abs(phase_delta) <= group_phase_tolerance) {
float group_score = score_group(&st->group, phase_delta);
if (st->matcher_cache[i].candidate_score < group_score) {
assert(i < DSSS_MATCHER_CACHE_SIZE);
/* Append to entry */
st->matcher_cache[i].candidate_score = group_score;
st->matcher_cache[i].candidate_phase = group_phase;
st->matcher_cache[i].candidate_data = decode_peak(st->group.max_ch, st->group.max);
st->matcher_cache[i].candidate_skips = 0;
}
}
/* Search for weakest entry */
float score = st->matcher_cache[i].last_score;
if (score < min_score) {
min_idx = i;
min_score = score;
}
}
/* If we found empty entries, replace one by a new decoding starting at this group */
if (empty_idx >= 0) {
assert(0 <= empty_idx && empty_idx < DSSS_MATCHER_CACHE_SIZE);
st->matcher_cache[empty_idx].last_phase = group_phase;
st->matcher_cache[empty_idx].candidate_score = base_score;
st->matcher_cache[empty_idx].last_score = base_score;
st->matcher_cache[empty_idx].candidate_phase = group_phase;
st->matcher_cache[empty_idx].candidate_data = decode_peak(st->group.max_ch, st->group.max);
st->matcher_cache[empty_idx].data_pos = 0;
st->matcher_cache[empty_idx].candidate_skips = 0;
st->matcher_cache[empty_idx].last_skips = 0;
/* If the weakest decoding in cache is weaker than a new decoding starting here, replace it */
} else if (min_score < base_score && min_idx >= 0) {
assert(0 <= min_idx && min_idx < DSSS_MATCHER_CACHE_SIZE);
st->matcher_cache[min_idx].last_phase = group_phase;
st->matcher_cache[min_idx].candidate_score = base_score;
st->matcher_cache[min_idx].last_score = base_score;
st->matcher_cache[min_idx].candidate_phase = group_phase;
st->matcher_cache[min_idx].candidate_data = decode_peak(st->group.max_ch, st->group.max);
st->matcher_cache[min_idx].data_pos = 0;
st->matcher_cache[min_idx].candidate_skips = 0;
st->matcher_cache[min_idx].last_skips = 0;
}
}
float run_iir(const float x, const int order, const struct iir_biquad q[order], struct iir_biquad_state st[order]) {
float intermediate = x;
for (int i=0; i<(order+1)/2; i++)
intermediate = run_biquad(intermediate, &q[i], &st[i]);
return intermediate;
}
float run_biquad(float x, const struct iir_biquad *const q, struct iir_biquad_state *const restrict st) {
/* direct form 2, see https://en.wikipedia.org/wiki/Digital_biquad_filter */
float intermediate = x + st->reg[0] * -q->a[0] + st->reg[1] * -q->a[1];
float out = intermediate * q->b[0] + st->reg[0] * q->b[1] + st->reg[1] * q->b[2];
st->reg[1] = st->reg[0];
st->reg[0] = intermediate;
return out;
}
float cwt_convolve_step(const float v[DSSS_WAVELET_LUT_SIZE], size_t offx) {
float sum = 0.0f;
for (ssize_t j=0; j<DSSS_WAVELET_LUT_SIZE; j++) {
/* Our wavelet is symmetric so convolution and correlation are identical. Use correlation here for ease of
* implementation */
sum += v[(offx + j) % DSSS_WAVELET_LUT_SIZE] * dsss_cwt_wavelet_table[j];
//DEBUG_PRINT(" j=%d v=%f w=%f", j, v[(offx + j) % DSSS_WAVELET_LUT_SIZE], dsss_cwt_wavelet_table[j]);
}
return sum;
}
/* Compute last element of correlation for input [a] and hard-coded gold sequences.
*
* This is intened to be used once for each new incoming sample in [a]. It expects [a] to be of length
* [dsss_correlation_length] and produces the one sample where both the reference sequence and the input fully overlap.
* This is equivalent to "valid" mode in numpy's terminology[0].
*
* [0] https://docs.scipy.org/doc/numpy/reference/generated/numpy.correlate.html
*/
float gold_correlate_step(const size_t ncode, const float a[DSSS_CORRELATION_LENGTH], size_t offx, bool debug) {
float acc_outer = 0.0f;
uint8_t table_byte = 0;
if (debug) DEBUG_PRINTN("Correlate n=%zd: ", ncode);
for (size_t i=0; i<DSSS_GOLD_CODE_LENGTH; i++) {
if ((i&7) == 0) {
table_byte = dsss_gold_code_table[ncode][i>>3]; /* Fetch sequence table item */
if (debug) DEBUG_PRINTN("|");
}
int bv = table_byte & (0x80>>(i&7)); /* Extract bit */
bv = !!bv*2 - 1; /* Map 0, 1 -> -1, 1 */
if (debug) DEBUG_PRINTN("%s%d\033[0m", bv == 1 ? "\033[92m" : "\033[91m", (bv+1)/2);
float acc_inner = 0.0f;
for (size_t j=0; j<DSSS_DECIMATION; j++)
acc_inner += a[(offx + i*DSSS_DECIMATION + j) % DSSS_CORRELATION_LENGTH]; /* Multiply item */
//if (debug) DEBUG_PRINTN("%.2f ", acc_inner);
acc_outer += acc_inner * bv;
}
if (debug) DEBUG_PRINTN("\n");
return acc_outer / DSSS_CORRELATION_LENGTH;
}
|