高通sensor resample分析

前言

最近做sensor二供料的兼容，发现，两个料的sensor sample rate不一致，会影响到算法的滤波器效果，因此需要做一下resample，借鉴了一下高通的resample方式，此篇文章做一个简单的分析

一、高通resample方式

高通resample的实现函数是在resampler/src/sns_resampler_sensor_instance_island.c中

static sns_rc
resampler_process_sample( resampler_instance_state *state,
                          resampler_sample * restrict input_sample,
                          resampler_sample * restrict output_sample, int32_t axis_cnt)
{
  sns_rc rc = SNS_RC_SUCCESS;

  //allocating the required space using uint64_t (so that the start address is aligned) and typecasting to resampler_sample
  uint64_t t_sample[((ALIGN_8(sizeof(resampler_sample)) + axis_cnt*sizeof(float))/sizeof(uint64_t))+1];
  resampler_sample* tmp_sample=(resampler_sample*)t_sample;
  tmp_sample->axis_cnt_bytes=axis_cnt * sizeof(float);
  tmp_sample->data=(float*)((uint8_t*)tmp_sample + ALIGN_8(sizeof(resampler_sample)));

  sns_time sample_timestamp = 0;
  bool new_sample = false;

  // PEND: sample sample check

  resampler_cic_obj_s* cic_obj = &state->cic_obj;
  resampler_interp_obj_s* int_obj = &state->interp_obj;

  sample_timestamp = input_sample->ts;

  switch ( state->quality )
  {
  case SNS_RESAMPLER_QUALITY_CURRENT_SAMPLE:
    if ( state->cur_samp_obj.sampling_count == 0 )
    {
      // PEND: add min_acceptable_gap check
      sns_memscpy(output_sample->data,output_sample->axis_cnt_bytes,
                  input_sample->data,input_sample->axis_cnt_bytes);
      output_sample->ts = sample_timestamp;
    }
    state->cur_samp_obj.sampling_count++;
    state->cur_samp_obj.sampling_count %= state->cur_samp_obj.sampling_factor;
    //assigning input sample accuracy to output sample
    output_sample->status = input_sample->status;
    break;

  case SNS_RESAMPLER_QUALITY_INTERPOLATED:
    resampler_interp_update(int_obj, input_sample, &output_sample->status);
    new_sample = resampler_interpolate(int_obj, output_sample->data, axis_cnt);
    if ( new_sample )
    {
      output_sample->ts = int_obj->desired_timestamp;
      RESAMPLER_UPDATE_INTERP_TS(int_obj);
    }
    break;

  case SNS_RESAMPLER_QUALITY_INTERPOLATED_FILTERED:
    resampler_interp_update(int_obj, input_sample, &output_sample->status);
    new_sample = resampler_interpolate(int_obj, tmp_sample->data, axis_cnt);

    if ( !new_sample )
    {
      break;
    }
    tmp_sample->ts = int_obj->desired_timestamp;
    RESAMPLER_UPDATE_INTERP_TS(int_obj);
    //Output sample has the intermediate accuracy status of the interpolation. This needs to be assigned to the tmp_sample before passing it on for filtering
    tmp_sample->status = output_sample->status;

    sns_memscpy(input_sample->data, input_sample->axis_cnt_bytes,
                tmp_sample->data, tmp_sample->axis_cnt_bytes);
    input_sample->ts     = tmp_sample->ts;
    input_sample->status = tmp_sample->status;

    /* fallthrough */
  case SNS_RESAMPLER_QUALITY_FILTERED:
    resampler_cic_update(cic_obj,
                         input_sample,
                         output_sample,
                         axis_cnt);
    break;
  }
  return rc;
}

可以看到，一共是分为了4种方式

    SNS_RESAMPLER_QUALITY_CURRENT_SAMPLE = 0,
    SNS_RESAMPLER_QUALITY_FILTERED = 1,
    SNS_RESAMPLER_QUALITY_INTERPOLATED_FILTERED = 2,
    SNS_RESAMPLER_QUALITY_INTERPOLATED = 3

1.1 简单抽值

SNS_RESAMPLER_QUALITY_CURRENT_SAMPLE
这是一种简单抽值的方式，适用于输入采样率和输出采样率是倍数关系，

1.2 线性插值

SNS_RESAMPLER_QUALITY_FILTERED
SNS_RESAMPLER_QUALITY_INTERPOLATED_FILTERED
两个都是线性插值，通过在相邻两个数据点之间进行线性插值来估计中间数值，
不过第二个带了滤波，会将输出的值，赋值给Input，用于下一次值的计算（没细看，猜的），

1.3 cic滤波器

积分梳状滤波器，一种比较经典的滤波器，常用于resample，以上的几种降采样方式，在高频信号中，可能会出现频谱混叠，一般都会做一个抗混叠滤波器（一般为低通滤波器，根据实际应用去选择，也可以用带通），最开始我们也是考虑的这种方式，但是会十分消耗硬件资源，因此放弃了。
因此CIC滤波器实现了一种简单的方式，可以规避这个问题。

二、结果

最终考虑到我们自己的sensor架构和硬件性能，选用了线性插值的方式做resample，