mlx9064x 0.2.1

Library for interfacing with Melexis MLX9064* thermal cameras
Documentation
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
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
// SPDX-License-Identifier: Apache-2.0
// Copyright © 2021 Will Ross
use core::iter;
use core::slice;

use arrayvec::ArrayVec;
use embedded_hal::blocking::i2c;

// Various floating point operations are not implemented in core, so we use libm to provide them as
// needed.
#[cfg_attr(feature = "std", allow(unused_imports))]
use num_traits::Float;

use crate::common::*;
use crate::error::{Error, LibraryError};
use crate::expose_member;
use crate::register::{AccessPattern, Subpage};
use crate::util::{i16_from_bits, Buffer};

use super::address::EepromAddress;
use super::Mlx90640;

/// The number of corner temperatures an MLX90640 has.
const NUM_CORNER_TEMPERATURES: usize = 4;

/// The word size of the MLX90640 in terms of 8-bit bytes.
const WORD_SIZE: usize = 16 / 8;

/// MLX90640-specific calibration processing.
#[derive(Clone, Debug, PartialEq)]
pub struct Mlx90640Calibration {
    k_v_dd: i16,

    v_dd_25: i16,

    resolution: u8,

    k_v_ptat: f32,

    k_t_ptat: f32,

    v_ptat_25: f32,

    alpha_ptat: f32,

    gain: f32,

    k_s_ta: f32,

    corner_temperatures: [i16; NUM_CORNER_TEMPERATURES],

    k_s_to: [f32; NUM_CORNER_TEMPERATURES],

    alpha_correction: [f32; NUM_CORNER_TEMPERATURES],

    alpha_pixels: [f32; <Self as CalibrationData>::Camera::NUM_PIXELS],

    alpha_cp: [f32; 2],

    offset_reference_pixels: [i16; <Self as CalibrationData>::Camera::NUM_PIXELS],

    offset_reference_cp: [i16; 2],

    k_v_pattern: [f32; 4],

    k_v_cp: f32,

    k_ta_pixels: [f32; <Self as CalibrationData>::Camera::NUM_PIXELS],

    k_ta_cp: f32,

    temperature_gradient_coefficient: Option<f32>,

    interleave_correction_pixels: [f32; 6],

    interleave_correction_cp: f32,
}

impl Mlx90640Calibration {
    /// Calculate pixel calibration values based off of the row and column data.
    ///
    /// Remainder data will be added in afterwards. This function is used for both offset and
    /// sensitivity (alpha) arrays. The given buffer must be at the word containing the column, row
    /// and remainder scaling factors.
    /// The calculated array, the remainder scaling factor, and the value occupying the 4 bits
    /// preceding the scaling factors are returned (in that order).
    fn calculate_bulk_pixel_calibration(
        data: &mut &[u8],
    ) -> ([i16; <Self as CalibrationData>::Camera::NUM_PIXELS], u8, u8) {
        let (extra_value, row_scale, column_scale, remainder_scale) = {
            let scales = word_to_u4s(data);
            (scales[0], scales[1], scales[2], scales[3])
        };
        let offset_average = data.get_i16();
        let mut pixel_calibration = [offset_average; Mlx90640::NUM_PIXELS];
        const VALUES_PER_DATA_ROW: usize = 4;
        // Add row offsets
        for row_chunks in pixel_calibration.chunks_exact_mut(Mlx90640::WIDTH * VALUES_PER_DATA_ROW)
        {
            let rows_coefficients = word_to_i4s(data);
            // Create a nice lazy iterator that converts the values to i16, scales them, and
            // reverses that order of the data (because the data is laid out backwards in the EEPROM).
            let rows_coefficients = core::array::IntoIter::new(rows_coefficients)
                .map(i16::from)
                .map(|coeff| coeff << row_scale)
                .rev();
            for (row, coefficient) in row_chunks
                .chunks_exact_mut(Mlx90640::WIDTH)
                .zip(rows_coefficients)
            {
                row.iter_mut().for_each(|element| *element += coefficient);
            }
        }
        // Add column offsets. Slightly more involved as the offsets are in row-major order.
        for column_chunk_index in 0..(Mlx90640::WIDTH / VALUES_PER_DATA_ROW) {
            // TODO: This could probably be optimized better
            let column_coefficients = word_to_i4s(data);
            // Same deal as the row coefficients, except cycle so that the same iterator can be
            // re-used multiple times.
            let column_coefficients = column_coefficients
                .iter()
                .copied()
                .map(i16::from)
                .map(|coeff| coeff << column_scale)
                .rev();
            for row in pixel_calibration.chunks_exact_mut(Mlx90640::WIDTH) {
                let start_index = column_chunk_index * VALUES_PER_DATA_ROW;
                let row_range = start_index..(start_index + VALUES_PER_DATA_ROW);
                row[row_range]
                    .iter_mut()
                    .zip(column_coefficients.clone())
                    .for_each(|(element, coefficient)| *element += coefficient);
            }
        }
        (pixel_calibration, remainder_scale, extra_value)
    }

    /// Generate a chessboard-patterned sequence from four values.
    ///
    /// The given values are ordered:
    /// 1. even row, even column
    /// 2. odd row, even column
    /// 3. even row, odd column
    /// 4. odd row, odd column
    ///
    /// This order is the same order the MLX90640 stores this calibration data in its EEPROM. Also
    /// note that the naming scheme in this library is 0-indexed, while the datasheet is 1-indexed,
    /// meaning even and odd are swapped between the two.
    ///
    /// The returned iterator yields values in row-major order.
    fn repeat_chessboard<T>(source_values: [i8; 4]) -> impl Iterator<Item = T>
    where
        T: From<i8> + core::fmt::Debug,
    {
        let row_even_col_even = source_values[0];
        let row_odd_col_even = source_values[1];
        let row_even_col_odd = source_values[2];
        let row_odd_col_odd = source_values[3];
        // Create a pattern for even or odd rows, starting from column 0
        let even_row_pattern = core::array::IntoIter::new([row_even_col_even, row_even_col_odd]);
        let odd_row_pattern = core::array::IntoIter::new([row_odd_col_even, row_odd_col_odd]);
        // Repeat the pattern across the row
        let even_row = even_row_pattern.cycle().take(Mlx90640::WIDTH).map(T::from);
        let odd_row = odd_row_pattern.cycle().take(Mlx90640::WIDTH).map(T::from);
        // Then chain the two rows together, repeating to fill the array
        let repeating_rows = even_row.chain(odd_row).cycle();
        repeating_rows.take(Mlx90640::NUM_PIXELS)
    }

    /// Calculate the per-pixel K<sub>T<sub>A</sub></sub> values.
    ///
    /// Similar to the offset and sensitivity values, the K<sub>T<sub>A</sub></sub> values have a
    /// per-pixel calibration value that is added to an average value shared by multiple pixels.
    /// Where the offset and sensitivity averages are calculated on a per row and column basis,
    /// this value is chosen from four values, determined by if the row and column indices are even
    /// or odd. The rest of the calculation is performed later, with the rest of the per-pixel
    /// calculations.
    fn generate_k_ta_pixels(data: &mut &[u8]) -> impl Iterator<Item = i16> {
        let source_data: ArrayVec<i8, 4> = (0..4).map(|_| data.get_i8()).collect();
        let source_data = source_data.into_inner().unwrap();
        Self::repeat_chessboard(source_data)
    }

    fn get_access_pattern_corrections(data: &mut &[u8]) -> (f32, [f32; 6]) {
        let word = data.get_u16();
        // The correction factors are 5 bits, 5 bits, then 6 bits, for factors 3, 2, 1.
        let chess_1_raw = i16_from_bits(&(word & 0x003F).to_be_bytes(), 6);
        let chess_2_raw = i16_from_bits(&((word & 0x07C0) >> 6).to_be_bytes(), 5);
        let chess_3_raw = i16_from_bits(&((word & 0xF800) >> 11).to_be_bytes(), 5);
        // factor 1 is scaled by 2^4
        let chess_1 = f32::from(chess_1_raw) / 16f32;
        // factor 2 is scaled by 2 (no exponent)
        let chess_2 = f32::from(chess_2_raw) / 2f32;
        // factor 3 is scaled by 2^3
        let chess_3 = f32::from(chess_3_raw) / 8f32;
        // See the docstring for InterleaveCorrectionIter for the explanation for this pattern
        let b = chess_2 - chess_3;
        let c = chess_2 + chess_3;
        let pattern = [-chess_3, b, -c, chess_3, -b, c];
        (chess_1, pattern)
    }

    /// Generate the constants needed for temperature calculations from a dump of the MLX90640
    /// EEPROM.
    ///
    /// The buffer must cover *all* of the EEPROM.
    pub fn from_data(data: &[u8]) -> Result<Self, LibraryError> {
        let mut buf = data;
        let eeprom_length = usize::from(EepromAddress::End - EepromAddress::Base);
        if buf.len() < eeprom_length {
            return Err(LibraryError::Other(
                "Not enough space left in buffer to be a full EEPROM dump",
            ));
        }
        // Skip the first 16 words, they're irrelevant
        buf.advance(WORD_SIZE * 16);
        // alpha_PTAT and offset compensation correction scales
        let (mut offset_reference_pixels, offset_correction_remainder_scale, alpha_ptat) =
            Self::calculate_bulk_pixel_calibration(&mut buf);
        let alpha_ptat = alpha_ptat / 4 + 8;
        let (alpha_pixels, alpha_correction_remainder_scale, alpha_scale_exp) =
            Self::calculate_bulk_pixel_calibration(&mut buf);
        let alpha_pixels: ArrayVec<f32, { Mlx90640::NUM_PIXELS }> =
            core::array::IntoIter::new(alpha_pixels)
                .map(f32::from)
                .collect();
        // Safe to unwrap as the length of pixel arrays are *all* Mlx90640::NUM_PIXELS long.
        let mut alpha_pixels = alpha_pixels.into_inner().unwrap();
        // Calculate the actual alpha scaling value from the exponent value. The alpha scaling
        // exponenet also has 30 added to it (not 27 like alpha_scale_cp).
        let alpha_scale = f32::from(alpha_scale_exp + 30).exp2();
        let gain = buf.get_i16();
        let v_ptat_25 = buf.get_i16();
        let (k_v_ptat, kt_ptat_bytes) = word_6_10_split(&mut buf);
        // k_v_ptat is scaled by 2^12
        let k_v_ptat = f32::from(k_v_ptat) / 4096f32;
        // k_t_ptat is scaled by 2^3
        let k_t_ptat = f32::from(i16_from_bits(&kt_ptat_bytes, 10)) / 8f32;
        let k_v_dd = (buf.get_i8() as i16) << 5;
        // The data in EEPROM is unsigned, so we upgrade to a signed type as it's immediately sent
        // negative (by subtracting 256), then multipled by 2^5, and finally has 2^13 subtracted
        // from it.
        let v_dd_25 = ((buf.get_u8() as i16) - 256) * (1 << 5) - (1 << 13);
        // Keep this value around for actual processing once we have kv_scale.
        let k_v_avg = word_to_i4s(&mut buf);
        let (interleave_correction_cp, interleave_correction_pixels) =
            Self::get_access_pattern_corrections(&mut buf);
        let lazy_k_ta_pixels = Self::generate_k_ta_pixels(&mut buf);
        let unpacked_scales = word_to_u4s(&mut buf);
        // The resolution control calibration value is just two bits in the high half of the byte.
        // The two other two bits are reserved, so we just drop them.
        let resolution = unpacked_scales[0] & 0x3;
        // various scaling constants
        let k_v_scale = f32::from(unpacked_scales[1]).exp2();
        // k_ta_scale1 has 8 added to it.
        let k_ta_scale1 = f32::from(unpacked_scales[2] + 8).exp2();
        // Leaving k_ta_scale2 as just the exponenet, as it's small enough to be used to shift it's
        // operand directly.
        let k_ta_scale2_exp = unpacked_scales[3];
        // We have k_v_scale now, calculate k_v_pattern
        let k_v_pattern: ArrayVec<f32, 4> = core::array::IntoIter::new(k_v_avg)
            .map(|v| f32::from(v) / k_v_scale)
            .collect();
        // Safe to unwrap as the input was only four elements, and the array is only 4 elements.
        let k_v_pattern = k_v_pattern.into_inner().unwrap();
        // Compensation pixel parameters
        let alpha_cp = {
            let (alpha_cp_ratio, alpha_cp_bytes) = word_6_10_split(&mut buf);
            let alpha_cp_ratio = f32::from(alpha_cp_ratio) / 7f32.exp2();
            // NOTE: the alpha scale value read from EEPROM has 27 added to get alpha_scale_cp, but
            // 30 added for alpha_scale_pixel
            let alpha_scale_cp = f32::from(alpha_scale_exp + 27).exp2();
            let alpha_cp0: f32 = f32::from(u16::from_be_bytes(alpha_cp_bytes)) / alpha_scale_cp;
            [alpha_cp0, alpha_cp0 * (1f32 + alpha_cp_ratio)]
        };
        let offset_reference_cp = {
            let (offset_cp_delta, offset_cp_bytes) = word_6_10_split(&mut buf);
            let offset_cp0 = i16_from_bits(&offset_cp_bytes, 10);
            [offset_cp0, offset_cp0 + i16::from(offset_cp_delta)]
        };
        let k_v_cp = f32::from(buf.get_i8()) / k_v_scale;
        let k_ta_cp = f32::from(buf.get_i8()) / k_ta_scale1;
        let k_s_ta = f32::from(buf.get_i8()) / 13f32.exp2();
        let temperature_gradient_coefficient = match buf.get_i8() {
            0 => None,
            n => Some(f32::from(n) / 5f32.exp2()),
        };
        // k_s_to is unscaled until k_s_to_scale is unpacked.
        let mut k_s_to_ranges: ArrayVec<f32, 4> = (0..4).map(|_| f32::from(buf.get_i8())).collect();
        // Fix the ordering of the elements from the EEPROM
        k_s_to_ranges.swap(0, 1);
        k_s_to_ranges.swap(2, 3);
        // Safe to unwrap as I'm just using ArrayVec to collect into an array.
        let mut k_s_to = k_s_to_ranges.into_inner().unwrap();
        // Very similar to the resolution and k_*_scale word a few lines above.
        let unpacked_corner_temps = word_to_u4s(&mut buf);
        // Like before, the top two bits are reserved. This time though, the temperature step
        // is multipled by 10. Also convert to i16 for use in calculations.
        let corner_temperature_step = i16::from(unpacked_corner_temps[0] & 0x3) * 10;
        // corner temperatures need to be multipled by the step and converted to i16
        let ct2 = i16::from(unpacked_corner_temps[2]) * corner_temperature_step;
        let ct3 = i16::from(unpacked_corner_temps[1]) * corner_temperature_step + ct2;
        // k_s_to_scale needs 8 added to it, then take 2 raised to this value.
        let k_s_to_scale = f32::from(unpacked_corner_temps[3] + 8).exp2();
        // -40 and 0 are hard-coded values for CT0 and CT1 (labelled CT1 and CT2 in the datasheet)
        let corner_temperatures = [-40i16, 0, ct2, ct3];
        // Now that we have k_s_to_scale, we can scale k_s_to properly:
        k_s_to.iter_mut().for_each(|k_s_to| *k_s_to /= k_s_to_scale);
        let basic_range = <Self as CalibrationData>::Camera::BASIC_TEMPERATURE_RANGE;
        let alpha_correction =
            alpha_correction_coefficients(basic_range, &corner_temperatures, &k_s_to);
        // Calculate the rest of the per-pixel data using the remainder/k_ta data
        let mut k_ta_pixels = [0f32; Mlx90640::NUM_PIXELS];
        offset_reference_pixels
            .iter_mut()
            .zip(alpha_pixels.iter_mut())
            .zip(lazy_k_ta_pixels)
            .zip(k_ta_pixels.iter_mut())
            .for_each(|(((offset, alpha), k_ta_rc), k_ta)| {
                // TODO: handle failed pixels (where the pixel data is 0x0000)
                // TODO: outlier/deviant pixels
                let high = buf.get_u8();
                let low = buf.get_u8();
                // Normal dance to extend the sign bit from an i6 to an i8
                let offset_remainder = i16::from(i8::from_ne_bytes([high & 0xFC]) >> 2);
                *offset += offset_remainder << offset_correction_remainder_scale;
                // alpha is going to be a little weird: not only is there the i6-shift-dance, but
                // there's an extra shift right by 4 to drop the k_ta and outlier bits.
                let alpha_remainder = (i16::from_be_bytes([high & 0x03, low]) << 6) >> 10;
                *alpha += f32::from(alpha_remainder << alpha_correction_remainder_scale);
                *alpha /= alpha_scale;
                // To try to keep floating point errors down as long as possible, do all the
                // operations for the numerator as ints, then convert to floats for the final division.
                let k_ta_remainder = i16::from(i8::from_ne_bytes([low & 0x0E]) << 4 >> 5);
                let k_ta_numerator = f32::from(k_ta_rc + (k_ta_remainder << k_ta_scale2_exp));
                *k_ta = k_ta_numerator / k_ta_scale1;
            });
        Ok(Self {
            k_v_dd,
            v_dd_25,
            resolution,
            k_v_ptat,
            k_t_ptat,
            v_ptat_25: v_ptat_25.into(),
            alpha_ptat: alpha_ptat.into(),
            gain: gain.into(),
            k_s_ta,
            corner_temperatures,
            k_s_to,
            alpha_correction,
            alpha_pixels,
            alpha_cp,
            offset_reference_pixels,
            offset_reference_cp,
            k_v_pattern,
            k_v_cp,
            k_ta_pixels,
            k_ta_cp,
            temperature_gradient_coefficient,
            interleave_correction_pixels,
            interleave_correction_cp,
        })
    }
}

impl<I2C> FromI2C<I2C> for Mlx90640Calibration
where
    I2C: i2c::WriteRead + i2c::Write,
{
    type Error = Error<I2C>;
    type Ok = Self;

    fn from_i2c(bus: &mut I2C, i2c_address: u8) -> Result<Self, Error<I2C>> {
        // Dump the EEPROM. Both cameras use the same size and starting offset for their EEPROM.
        const EEPROM_LENGTH: usize =
            (EepromAddress::End as usize - EepromAddress::Base as usize + 1) * 2;
        let mut eeprom_buf = [0u8; EEPROM_LENGTH];
        let eeprom_base: Address = EepromAddress::Base.into();
        bus.write_read(i2c_address, &eeprom_base.as_bytes(), &mut eeprom_buf)
            .map_err(Error::I2cWriteReadError)?;
        Ok(Self::from_data(&eeprom_buf)?)
    }
}

impl<'a> CalibrationData<'a> for Mlx90640Calibration {
    type Camera = Mlx90640;

    expose_member!(k_v_dd, i16);
    expose_member!(v_dd_25, i16);
    expose_member!(resolution, u8);
    expose_member!(k_v_ptat, f32);
    expose_member!(k_t_ptat, f32);
    expose_member!(v_ptat_25, f32);
    expose_member!(alpha_ptat, f32);
    expose_member!(gain, f32);
    expose_member!(k_s_ta, f32);

    expose_member!(&corner_temperatures, [i16]);
    expose_member!(&k_s_to, [f32]);
    expose_member!(&alpha_correction, [f32]);

    type OffsetReferenceIterator = slice::Iter<'a, i16>;

    fn offset_reference_pixels(&'a self, _subpage: Subpage) -> Self::OffsetReferenceIterator {
        self.offset_reference_pixels.iter()
    }

    fn offset_reference_cp(&self, subpage: Subpage) -> i16 {
        self.offset_reference_cp[subpage as usize]
    }

    type AlphaIterator = slice::Iter<'a, f32>;

    fn alpha_pixels(&'a self, _subpage: Subpage) -> Self::AlphaIterator {
        self.alpha_pixels.iter()
    }

    fn alpha_cp(&self, subpage: Subpage) -> f32 {
        self.alpha_cp[subpage as usize]
    }

    type KvIterator = ChessboardIter<'a, f32>;

    fn k_v_pixels(&'a self, _subpage: Subpage) -> Self::KvIterator {
        ChessboardIter::new(&self.k_v_pattern)
    }

    fn k_v_cp(&self, _subpage: Subpage) -> f32 {
        self.k_v_cp
    }

    type KtaIterator = slice::Iter<'a, f32>;

    fn k_ta_pixels(&'a self, _subpage: Subpage) -> Self::KtaIterator {
        self.k_ta_pixels.iter()
    }

    fn k_ta_cp(&self, _subpage: Subpage) -> f32 {
        self.k_ta_cp
    }

    expose_member!(temperature_gradient_coefficient, Option<f32>);

    type AccessPatternCompensation = PixelAccessPatternCompensation<'a>;

    /// A sequence of per-pixel correction values that are added to the pixl gain value.
    ///
    /// The MLX90640 can be used in interleaved mode, but for optimal performance a correction
    /// needs to be applied. This value is summed with the pixel gain value and reference offset
    /// (the reference offset being scaled relative to the temperature difference).
    fn access_pattern_compensation_pixels(
        &'a self,
        access_pattern: AccessPattern,
    ) -> Self::AccessPatternCompensation {
        PixelAccessPatternCompensation::new(access_pattern, &self.interleave_correction_pixels[..])
    }

    /// Equivalent to [`Self::access_pattern_compensation_pixels`] for compensation pixels.
    fn access_pattern_compensation_cp(
        &self,
        subpage: Subpage,
        access_pattern: AccessPattern,
    ) -> Option<f32> {
        match (subpage, access_pattern) {
            (Subpage::One, AccessPattern::Interleave) => Some(self.interleave_correction_cp),
            _ => None,
        }
    }
}

#[derive(Clone, Debug, PartialEq)]
pub struct ChessboardIter<'a, T: 'a> {
    index: usize,
    source: &'a [T],
}

impl<'a, T: 'a> ChessboardIter<'a, T> {
    /// Repeat a sequence of values in a chessboard pattern.
    ///
    /// The given slice must have at least four values (if it has more than four, they are
    /// ignored). The values in the slice will be used for values like so (assuming 0-indexed rows
    /// and columns):
    /// 1. even row, even column
    /// 2. odd row, even column
    /// 3. even row, odd column
    /// 4. odd row, odd column
    fn new(source: &'a [T]) -> Self {
        Self { index: 0, source }
    }
}

impl<'a, T: 'a> Iterator for ChessboardIter<'a, T> {
    type Item = &'a T;

    fn next(&mut self) -> Option<Self::Item> {
        if self.index < Mlx90640::NUM_PIXELS {
            let row = self.index / Mlx90640::WIDTH;
            let column = self.index % Mlx90640::WIDTH;
            self.index += 1;
            Some(match (row % 2 == 0, column % 2 == 0) {
                (true, true) => &self.source[0],
                (false, true) => &self.source[1],
                (true, false) => &self.source[2],
                (false, false) => &self.source[3],
            })
        } else {
            None
        }
    }
}

/// An iterator for the interleaved reading pattern values
///
#[doc = include_str!("../katex.html")]
///
/// These values are only used if the interleaved access mode is being active. It's described in
/// section 11.1.3.1 of the MLX90640 datasheet, but the formula is overly complicated:
///
/// $$
/// \begin{split}
/// IL\_{\textit{Pattern}} =& {} int\left(\frac{\mathtt{pixel\\_index} - 1}{32}\right) -
/// int\left(\frac{int\left(\frac{\mathtt{pixel\\_index} - 1}{32}\right)}{2}\right) \* 2 \newline
/// \textit{Conversion}\_{\textit{pattern}} =& \left(int\left(\frac{\mathtt{pixel\\_index} - 3}{4}\right)\right. \newline
/// &\qquad {} - int\left(\frac{\mathtt{pixel\\_index} - 2}{4}\right) \newline
/// &\qquad {} + int\left(\frac{\mathtt{pixel\\_index}}{4}\right) \newline
/// &\qquad \left. {} - int\left(\frac{\mathtt{pixel\\_index} - 1}{4}\right)
/// \right) * (1 - 2 * IL\_{\textit{Pattern}})
/// \end{split}
/// $$
///
/// Can be simplified to
///
/// $$
/// \begin{split}
/// IL\_{\textit{Pattern}} =& {} \mathtt{pixel\\_index} \bmod 2 \newline
/// \textit{Conversion}\_{\textit{pattern}} =& \left(\mathtt{pixel\\_index} \bmod 2\right) \newline
/// &\qquad {} * \left(\left(\mathtt{pixel\\_index} \bmod 4 \right) - 2\right) \newline
/// &\qquad {} * \left(1 - \left(int\left(\frac{\mathtt{pixel\\_index}}{32}\right) \bmod 2 * 2\right)\right)
/// \end{split}
/// $$
///
/// Those values are used for the actual correction formula:
///
/// \begin{align*}
/// \text{pix}\_{OS(i, j)} &= pix\_{gain(i, j)} \newline
/// &\qquad \colorbox{yellow}{$ {} + IL\_{CHESS\_3} \* \left(2 * IL\_{\textit{Pattern}} - 1 \right)$} \newline
/// &\qquad \colorbox{yellow}{${} - IL\_{CHESS\_2} \* \textit{Conversion}\_{\textit{pattern}}$} \newline
/// &\qquad {} - \textit{offset}\_{(i, j)} \newline
/// &\qquad {} \* (1 + K_{T\_{a}(i, j)} \* (T_a - T_{a_0})) \newline
/// &\qquad {} \* (1 + K_{V(i, j)} \* (V_{DD} - V_{DD_0}))
/// \end{align*}
///
/// This isn't stated in the datasheet, but if you actually calculate the values from
/// section 11.2.3.1, you see that there are two, four element patterns. The patterns are
/// alternated for each row, and the elements are repeated across the row.
/// ```text
/// -A  B -A -C
///  A -B  A  C
/// ```
///
/// Where $A = IL\_{CHESS\_3}$, $B = IL\_{CHESS\_2} - IL\_{CHESS\_3}$, and $C = IL\_{CHESS\_2} + IL\_{CHESS\_3}$
#[derive(Clone, Debug)]
pub enum PixelAccessPatternCompensation<'a> {
    Chess(iter::Take<iter::Repeat<Option<&'a f32>>>),
    Interleave {
        index: usize,
        /// The pattern values, split across two arrays of three elements
        patterns: &'a [f32],
    },
}

impl<'a> PixelAccessPatternCompensation<'a> {
    /// Create a new interleaved correction iterator from a slice of correction constants
    ///
    /// The slice must be 6 elements long, with the values laid out as `[-A, B, -C, A, -B, C]`.
    fn new(access_pattern: AccessPattern, patterns: &'a [f32]) -> Self {
        match access_pattern {
            AccessPattern::Chess => Self::Chess(iter::repeat(None).take(Mlx90640::NUM_PIXELS)),
            AccessPattern::Interleave => Self::Interleave { index: 0, patterns },
        }
    }
}

impl<'a> Iterator for PixelAccessPatternCompensation<'a> {
    type Item = Option<&'a f32>;

    fn next(&mut self) -> Option<Self::Item> {
        match self {
            PixelAccessPatternCompensation::Chess(inner) => inner.next(),
            PixelAccessPatternCompensation::Interleave { index, patterns } => {
                if *index < Mlx90640::NUM_PIXELS {
                    // We just need the even/odd-ness of the row index
                    let row_sign = (*index / Mlx90640::WIDTH) % 2;
                    let column_index = match *index % 4 {
                        0 | 2 => 0,
                        1 => 1,
                        3 => 2,
                        _ => unreachable!(),
                    };
                    let pattern_index = row_sign * 3 + column_index;
                    *index += 1;
                    Some(Some(&patterns[pattern_index]))
                } else {
                    None
                }
            }
        }
    }
}

/// Split a word into a 6-bit value and a 10-bit value.
///
/// Further conversion for the second value is left to the caller.
fn word_6_10_split(data: &mut &[u8]) -> (i8, [u8; 2]) {
    let mut word = [data.get_u8(), data.get_u8()];
    // Copy out the 6-bit value, and shift it over. As signed right shifts are aritmetic, the sign
    // bit gets extended, and we get the value we wanted.
    let six_bit = i8::from_ne_bytes([word[0]]) >> 2;
    // Mask off the extra from the high byte
    word[0] &= 0x03;
    (six_bit, word)
}

/// Extract four unsigned, 4-bit integers from a buffer
fn word_to_u4s(data: &mut &[u8]) -> [u8; 4] {
    let high = data.get_u8();
    let low = data.get_u8();
    [(high & 0xF0) >> 4, high & 0xF, (low & 0xF0) >> 4, low & 0xF]
}

/// Split a byte into a pair of signed, four-bit integers.
fn u8_to_i4s(byte: u8) -> [i8; 2] {
    // Start by splitting the two numbers into a pair of bytes, with the MSB of the i4 all the way
    // to the left.
    let high = byte & 0xF0;
    let low = (byte & 0xF) << 4;
    // Create i8s from the pair of bytes, then arithmetic shift right by 4 to extend the sign.
    // Endian-ness shouldn't matter as these are single bytes.
    let high = i8::from_ne_bytes([high]);
    let low = i8::from_ne_bytes([low]);
    [high >> 4, low >> 4]
}

/// Split a word (2 bytes) from a buffer into four, 4-bit signed integers.
fn word_to_i4s(data: &mut &[u8]) -> [i8; 4] {
    let high = data.get_u8();
    let low = data.get_u8();
    let high = u8_to_i4s(high);
    let low = u8_to_i4s(low);
    [high[0], high[1], low[0], low[1]]
}

#[cfg(test)]
#[allow(clippy::excessive_precision)]
pub(crate) mod test {
    #[cfg(feature = "std")]
    extern crate std;
    #[cfg(feature = "std")]
    use std::{print, println};

    use arrayvec::ArrayVec;
    use float_cmp::{assert_approx_eq, ApproxEq};

    use crate::common::{CalibrationData, MelexisCamera};
    use crate::mlx90640::Mlx90640;
    use crate::register::{AccessPattern, Subpage};
    use crate::test::{mlx90640_datasheet_eeprom, mlx90640_example_data};

    use super::Mlx90640Calibration;

    fn datasheet_eeprom() -> Mlx90640Calibration {
        let eeprom_bytes = mlx90640_datasheet_eeprom();
        Mlx90640Calibration::from_data(&eeprom_bytes).expect("The EEPROM data to be parsed.")
    }

    fn example_eeprom() -> Mlx90640Calibration {
        Mlx90640Calibration::from_data(mlx90640_example_data::EEPROM_DATA)
            .expect("The example data should be parseable")
    }

    #[test]
    fn word_6_10_split() {
        fn check(mut data: &[u8], little: i8, remainder: [u8; 2]) {
            let split = super::word_6_10_split(&mut data);
            assert_eq!(
                split.0, little,
                "word_6_10_split failed to split the upper 6 bits off"
            );
            assert_eq!(
                split.1, remainder,
                "word_6_10_split failed to split the lower 10 bits off"
            );
        }
        // Start with basic premise, that the first 6 bits are split off
        check(b"\xfc\x00", -1, [0x00, 0x00]);
        // Then check the "reverse"
        check(b"\x03\xff", 0, [0x03, 0xFF]);
        // Negative max i6, with u10::MAX as well
        check(b"\x83\xff", -32, [0x03, 0xFF]);
    }

    #[test]
    fn word_to_u4s() {
        let mut sequence: &[u8] = b"\x12\x34";
        assert_eq!(super::word_to_u4s(&mut sequence), [1, 2, 3, 4]);
        let mut max_min: &[u8] = b"\xf0\xf0";
        assert_eq!(super::word_to_u4s(&mut max_min), [0xF, 0, 0xF, 0]);
        let mut min_max: &[u8] = b"\x0f\x0f";
        assert_eq!(super::word_to_u4s(&mut min_max), [0, 0xF, 0, 0xF]);
    }

    #[test]
    fn u8_to_i4s() {
        assert_eq!(super::u8_to_i4s(0x44), [4, 4]);
        assert_eq!(super::u8_to_i4s(0x88), [-8, -8]);
        assert_eq!(super::u8_to_i4s(0x48), [4, -8]);
        assert_eq!(super::u8_to_i4s(0x84), [-8, 4]);
    }

    #[test]
    fn repeat_chessboard() {
        // The pattern order is (for row, column): EE, OE, EO, OO
        let pattern = [1, 2, 3, 4];
        let test_pattern: ArrayVec<i8, { Mlx90640::NUM_PIXELS }> =
            Mlx90640Calibration::repeat_chessboard(pattern).collect();
        // Print the test pattern (when std is available), as that makes it much easier to see
        // what's going on.
        #[cfg(feature = "std")]
        for row in 0..Mlx90640::HEIGHT {
            for column in 0..Mlx90640::WIDTH {
                let index = row * Mlx90640::WIDTH + column;
                print!("{} ", test_pattern[index]);
            }
            println!();
        }
        for column in 0..Mlx90640::WIDTH {
            for row in 0..Mlx90640::HEIGHT {
                let index = row * Mlx90640::WIDTH + column;
                let expected = match (row % 2, column % 2) {
                    (0, 0) => 1,
                    (1, 0) => 2,
                    (0, 1) => 3,
                    (1, 1) => 4,
                    (_, _) => unreachable!(),
                };
                assert_eq!(
                    test_pattern[index], expected,
                    "pattern incorrect at index {}",
                    index
                );
            }
        }
    }

    /// Check that it can even create itself from a buffer.
    #[test]
    fn smoke() {
        datasheet_eeprom();
        example_eeprom();
    }

    // Ordering these tests in the same order as the data sheet's worked example.
    #[test]
    fn resolution() {
        assert_eq!(datasheet_eeprom().resolution(), 2);
        assert_eq!(
            example_eeprom().resolution(),
            mlx90640_example_data::RESOLUTION_EE
        );
    }

    #[test]
    fn k_v_dd() {
        assert_eq!(datasheet_eeprom().k_v_dd(), -3168);
        assert_eq!(example_eeprom().k_v_dd(), mlx90640_example_data::K_V_DD);
    }

    #[test]
    fn v_dd_25() {
        assert_eq!(datasheet_eeprom().v_dd_25(), -13056);
        assert_eq!(example_eeprom().v_dd_25(), mlx90640_example_data::V_DD_25);
    }

    #[test]
    fn v_dd_0() {
        assert_eq!(datasheet_eeprom().v_dd_0(), 3.3);
        assert_eq!(example_eeprom().v_dd_0(), 3.3);
    }

    #[test]
    fn k_v_ptat() {
        assert_approx_eq!(
            f32,
            datasheet_eeprom().k_v_ptat(),
            0.005371094,
            epsilon = 0.000000001
        );
        assert_approx_eq!(
            f32,
            example_eeprom().k_v_ptat(),
            mlx90640_example_data::K_V_PTAT,
            epsilon = 0.00001
        );
    }

    #[test]
    fn k_t_ptat() {
        // These values are scaled by 1/8 and are not too large so they can be exactly represented
        // in an f32.
        assert_eq!(datasheet_eeprom().k_t_ptat(), 42.25);
        assert_eq!(example_eeprom().k_t_ptat(), mlx90640_example_data::K_T_PTAT);
    }

    #[test]
    fn v_ptat_25() {
        // These values are integers for the 640.
        assert_eq!(datasheet_eeprom().v_ptat_25(), 12273f32);
        assert_eq!(
            example_eeprom().v_ptat_25(),
            mlx90640_example_data::V_PTAT_25
        );
    }

    #[test]
    fn alpha_ptat() {
        // The example values are both integers
        assert_eq!(datasheet_eeprom().alpha_ptat(), 9f32);
        assert_eq!(
            example_eeprom().alpha_ptat(),
            mlx90640_example_data::ALPHA_PTAT
        );
    }
    #[test]

    fn gain() {
        // The EEPROM gain value is an integer for the 640
        assert_eq!(datasheet_eeprom().gain(), 6383f32);
        assert_eq!(example_eeprom().gain(), mlx90640_example_data::GAIN_EE);
    }

    fn test_pixels_common<T: PartialEq + core::fmt::Debug + core::fmt::Display + Copy>(
        datasheet_data: [ArrayVec<T, { Mlx90640::NUM_PIXELS }>; 2],
        example_data: [ArrayVec<T, { Mlx90640::NUM_PIXELS }>; 2],
        datasheet_expected: T,
        example_expected: &[T; Mlx90640::NUM_PIXELS],
        subpage: Option<Subpage>,
        check: &dyn Fn(T, T) -> bool,
    ) {
        if subpage.is_none() {
            assert_eq!(datasheet_data[0], datasheet_data[1]);
            assert_eq!(example_data[0], example_data[1]);
        }
        // Test the single pixel from the datasheet
        let datasheet_index = 11 * Mlx90640::WIDTH + 15;
        let subpage_index: usize = subpage.unwrap_or(Subpage::Zero).into();
        let pixel = datasheet_data[subpage_index][datasheet_index];
        assert!(
            check(pixel, datasheet_expected),
            "[datasheet pixel]:\n{:>10}: `{:?}`,\n{:>10}: `{:?}`,",
            "expected",
            datasheet_expected,
            "actual",
            pixel
        );
        // Check all the pixels from the full example
        let offset_pairs = example_data[subpage_index]
            .iter()
            .zip(example_expected.iter());
        for (index, (actual, expected)) in offset_pairs.enumerate() {
            assert!(
                check(*actual, *expected),
                "[pixel {:?}]:\n{:>10}: `{:?}`,\n{:>10}: `{:?}`,",
                index,
                "expected",
                expected,
                "actual",
                actual
            );
        }
    }

    fn test_pixels_approx<T>(
        datasheet_data: [ArrayVec<T, { Mlx90640::NUM_PIXELS }>; 2],
        example_data: [ArrayVec<T, { Mlx90640::NUM_PIXELS }>; 2],
        datasheet_expected: T,
        example_expected: &[T; Mlx90640::NUM_PIXELS],
        subpage: Option<Subpage>,
        margin: Option<<T as ApproxEq>::Margin>,
    ) where
        T: ApproxEq + PartialEq + core::fmt::Debug + core::fmt::Display + Copy,
    {
        let check = |actual: T, expected: T| actual.approx_eq(expected, margin.unwrap_or_default());
        test_pixels_common(
            datasheet_data,
            example_data,
            datasheet_expected,
            example_expected,
            subpage,
            &check,
        );
    }

    fn test_pixels<T: PartialEq + core::fmt::Debug + core::fmt::Display + Copy>(
        datasheet_data: [ArrayVec<T, { Mlx90640::NUM_PIXELS }>; 2],
        example_data: [ArrayVec<T, { Mlx90640::NUM_PIXELS }>; 2],
        datasheet_expected: T,
        example_expected: &[T; Mlx90640::NUM_PIXELS],
        subpage: Option<Subpage>,
    ) {
        let check = |actual: T, expected: T| actual == expected;
        test_pixels_common(
            datasheet_data,
            example_data,
            datasheet_expected,
            example_expected,
            subpage,
            &check,
        );
    }

    #[test]
    fn pixel_offset() {
        let datasheet = datasheet_eeprom();
        let datasheet_offsets: [ArrayVec<i16, { Mlx90640::NUM_PIXELS }>; 2] = [
            datasheet
                .offset_reference_pixels(Subpage::Zero)
                .copied()
                .collect(),
            datasheet
                .offset_reference_pixels(Subpage::One)
                .copied()
                .collect(),
        ];
        let example = example_eeprom();
        let example_offsets: [ArrayVec<i16, { Mlx90640::NUM_PIXELS }>; 2] = [
            example
                .offset_reference_pixels(Subpage::Zero)
                .copied()
                .collect(),
            example
                .offset_reference_pixels(Subpage::One)
                .copied()
                .collect(),
        ];
        test_pixels(
            datasheet_offsets,
            example_offsets,
            -75,
            &mlx90640_example_data::OFFSET_REFERENCE_PIXELS,
            // MLX90640 doesn't vary offsets on subpage
            None,
        );
    }

    #[test]
    fn pixel_k_ta() {
        let datasheet = datasheet_eeprom();
        let datasheet_k_ta: [ArrayVec<f32, { Mlx90640::NUM_PIXELS }>; 2] = [
            datasheet.k_ta_pixels(Subpage::Zero).copied().collect(),
            datasheet.k_ta_pixels(Subpage::One).copied().collect(),
        ];
        let example = example_eeprom();
        let example_k_ta: [ArrayVec<f32, { Mlx90640::NUM_PIXELS }>; 2] = [
            example.k_ta_pixels(Subpage::Zero).copied().collect(),
            example.k_ta_pixels(Subpage::One).copied().collect(),
        ];
        test_pixels_approx(
            datasheet_k_ta,
            example_k_ta,
            0.005126953125,
            &mlx90640_example_data::K_TA_PIXELS,
            // MLX90640 doesn't vary k_ta on subpage
            None,
            // The example spreadsheet goes out to eight decimal places
            Some((10E-8, 2).into()),
        );
    }

    #[test]
    fn k_v_pixels() {
        let datasheet = datasheet_eeprom();
        let datasheet_k_v: [ArrayVec<f32, { Mlx90640::NUM_PIXELS }>; 2] = [
            datasheet.k_v_pixels(Subpage::Zero).copied().collect(),
            datasheet.k_v_pixels(Subpage::One).copied().collect(),
        ];
        let example = example_eeprom();
        let example_k_v: [ArrayVec<f32, { Mlx90640::NUM_PIXELS }>; 2] = [
            example.k_v_pixels(Subpage::Zero).copied().collect(),
            example.k_v_pixels(Subpage::One).copied().collect(),
        ];
        test_pixels_approx(
            datasheet_k_v,
            example_k_v,
            0.5f32,
            &mlx90640_example_data::K_V_PIXELS,
            // MLX90640 doesn't vary k_v on subpage
            None,
            // The values in the datasheet are all exactly representable in a float, so no need for
            // an explicit margin.
            None,
        );
    }

    #[test]
    fn emissivity() {
        assert_eq!(datasheet_eeprom().emissivity(), None);
        assert_eq!(example_eeprom().emissivity(), None);
    }

    #[test]
    fn offset_reference_cp() {
        // datasheet
        let datasheet = datasheet_eeprom();
        assert_eq!(datasheet.offset_reference_cp(Subpage::Zero), -75);
        assert_eq!(datasheet.offset_reference_cp(Subpage::One), -77);
        // example
        let example = example_eeprom();
        assert_eq!(
            example.offset_reference_cp(Subpage::Zero),
            mlx90640_example_data::OFFSET_CP[0],
        );
        assert_eq!(
            example.offset_reference_cp(Subpage::One),
            mlx90640_example_data::OFFSET_CP[1],
        );
    }

    #[test]
    fn k_ta_cp() {
        // The 640 doesn't vary k_ta_cp on subpage.
        // datasheet
        let datasheet = datasheet_eeprom();
        // This value is more precise than an f32, so skipping an explicit margin
        let expected = 0.00457763671875;
        assert_approx_eq!(f32, datasheet.k_ta_cp(Subpage::Zero), expected);
        assert_approx_eq!(f32, datasheet.k_ta_cp(Subpage::One), expected);
        // example
        let example = example_eeprom();
        // The example value is only given to 6 decimal places
        assert_approx_eq!(
            f32,
            example.k_ta_cp(Subpage::Zero),
            mlx90640_example_data::K_TA_CP,
            epsilon = 10E-6
        );
        assert_approx_eq!(
            f32,
            example.k_ta_cp(Subpage::One),
            mlx90640_example_data::K_TA_CP,
            epsilon = 10E-6
        );
    }

    #[test]
    fn k_v_cp() {
        // Both the datasheet and the example spreadsheet values are fractional powers of 2 and can
        // be exactly represented by floats.
        // datasheet
        let datasheet = datasheet_eeprom();
        let expected = 0.5;
        assert_eq!(datasheet.k_v_cp(Subpage::Zero), expected);
        assert_eq!(datasheet.k_v_cp(Subpage::One), expected);
        // example
        let example = example_eeprom();
        assert_eq!(example.k_v_cp(Subpage::Zero), mlx90640_example_data::K_V_CP);
        assert_eq!(example.k_v_cp(Subpage::One), mlx90640_example_data::K_V_CP);
    }

    #[test]
    fn temperature_gradient_coefficient() {
        assert_eq!(
            datasheet_eeprom().temperature_gradient_coefficient(),
            Some(1f32)
        );
        assert_eq!(
            example_eeprom().temperature_gradient_coefficient(),
            // 0 TGC means None
            None,
        );
    }

    #[test]
    fn alpha_cp() {
        // datasheet
        let datasheet = datasheet_eeprom();
        // The datasheet values go out to 11 decimal places
        assert_approx_eq!(
            f32,
            datasheet.alpha_cp(Subpage::Zero),
            4.07453626394272E-9,
            epsilon = 10E-11
        );
        assert_approx_eq!(
            f32,
            datasheet.alpha_cp(Subpage::One),
            3.851710062200835E-9,
            epsilon = 10E-11
        );
        // example
        let example = example_eeprom();
        // The example data goes out to 19 decimal places (but only 12 significant figures).
        assert_approx_eq!(
            f32,
            example.alpha_cp(Subpage::Zero),
            mlx90640_example_data::ALPHA_CP[0],
            epsilon = 10E-19
        );
        assert_approx_eq!(
            f32,
            example.alpha_cp(Subpage::One),
            mlx90640_example_data::ALPHA_CP[1],
            epsilon = 10E-19
        );
    }

    #[test]
    fn k_s_ta() {
        assert_approx_eq!(
            f32,
            datasheet_eeprom().k_s_ta(),
            -0.001953125,
            // The datasheet value is specified out to 9 decimal places
            epsilon = 10E-9
        );
        assert_approx_eq!(
            f32,
            example_eeprom().k_s_ta(),
            mlx90640_example_data::K_S_TA,
            // Example value is only specified out to 6 decimal places
            epsilon = 10E-6
        );
    }

    #[test]
    fn pixel_alpha() {
        // MLX90640 doesn't vary alpha on subpage
        let datasheet = datasheet_eeprom();
        let datasheet_alpha: [ArrayVec<f32, { Mlx90640::NUM_PIXELS }>; 2] = [
            datasheet.alpha_pixels(Subpage::Zero).copied().collect(),
            datasheet.alpha_pixels(Subpage::One).copied().collect(),
        ];
        let example = example_eeprom();
        let example_alpha: [ArrayVec<f32, { Mlx90640::NUM_PIXELS }>; 2] = [
            example.alpha_pixels(Subpage::Zero).copied().collect(),
            example.alpha_pixels(Subpage::One).copied().collect(),
        ];
        test_pixels_approx(
            datasheet_alpha,
            example_alpha,
            1.262233122690854E-7,
            &mlx90640_example_data::ALPHA_PIXELS,
            None,
            // The example data goes out to 19 decimal places (but only 12 significant figures).
            Some((10E-19, 2).into()),
        );
    }

    #[test]
    fn k_s_to() {
        assert_approx_eq!(
            f32,
            datasheet_eeprom().k_s_to()[1],
            -0.00080108642578125,
            // Datasheet value is exactly reproduced above, so 17 decimal places
            epsilon = 10E-17
        );
        let example = example_eeprom();
        let example_pairs = example
            .k_s_to()
            .iter()
            .zip(mlx90640_example_data::K_S_TO.iter());
        for (actual, expected) in example_pairs {
            assert_approx_eq!(
                f32,
                *actual,
                *expected,
                // The example values are only specified out to 6 decimal places
                epsilon = 10E-6
            );
        }
    }

    #[test]
    fn corner_temperatures() {
        let e = datasheet_eeprom();
        let ct = e.corner_temperatures();
        // The first two values are hard-coded, but testing for completeness.
        assert_eq!(ct[0], -40);
        assert_eq!(ct[1], 0);
        // These are loaded from EEPROM
        assert_eq!(ct[2], 160);
        assert_eq!(ct[3], 320);

        // Full example test
        assert_eq!(
            example_eeprom().corner_temperatures(),
            mlx90640_example_data::CORNER_TEMPERATURES
        );
    }

    #[test]
    fn access_pattern_compensation_chess() {
        let e = datasheet_eeprom();
        for compensation in e.access_pattern_compensation_pixels(AccessPattern::Chess) {
            assert!(
                compensation.is_none(),
                "There is no access pattern compensation in chess mode"
            );
        }
    }

    #[test]
    fn access_pattern_compensation_interleave() {
        let e = datasheet_eeprom();
        // Using the variable names described in the `PixelAccessPatternCompensation` docstring.
        const IL_CHESS_3: f32 = 0.125;
        const IL_CHESS_2: f32 = 3.0;
        const A: f32 = IL_CHESS_3;
        const B: f32 = IL_CHESS_2 - IL_CHESS_3;
        const C: f32 = IL_CHESS_2 + IL_CHESS_3;
        let compensation_iter = e
            .access_pattern_compensation_pixels(AccessPattern::Interleave)
            .enumerate();
        let mut count = 0;
        for (pixel_index, compensation) in compensation_iter {
            let row = pixel_index / Mlx90640::WIDTH;
            let column = pixel_index % Mlx90640::WIDTH;
            // Start with the expected values for an even row
            let expected = if column % 2 == 0 {
                -A
            } else if column % 4 == 3 {
                -C
            } else {
                B
            };
            // Negate the value for an odd row
            let expected = if row % 2 == 0 { expected } else { -expected };
            assert_eq!(
                Some(&expected),
                compensation,
                "[pixel {:?}]:\n{:>10}: `{:?}`,\n{:>10}: `{:?}`,",
                pixel_index,
                "expected",
                Some(expected),
                "actual",
                compensation
            );
            count += 1;
        }
        assert_eq!(count, Mlx90640::NUM_PIXELS);
    }
}