Merge pull request #248 from anarkiwi/avg

I/Q inference gates on average power.

grc/iqtlabs_iq_inference.block.yml

@@ -25,7 +25,7 @@ documentation: |-
     vlen: length of complex/FFT dB vector.
     n_vlen: number of vectors aggregate.
     sample_buffer: size of sample lookback buffer.
-    min_peak_points: Only run inference on buckets with this minimum dB power.
+    min_peak_points: Only run inference with this minimum average dB power.
     model_names: if not empty, comma separated list of model names.
     model_server: if not empty, do torchserve inference to this address.
     confidence: Only output inference results where confidence is greater.

lib/iq_inference_impl.cc

@@ -250,8 +250,8 @@ iq_inference_impl::iq_inference_impl(const std::string &tag, COUNT_T vlen,
   batch_ = vlen_ * n_vlen_;
   samples_lookback_.reset(new gr_complex[batch_ * sample_buffer]);
   unsigned int alignment = volk_get_alignment();
-  max_.reset((uint16_t *)volk_malloc(sizeof(uint16_t), alignment));
-  total_.reset((float *)volk_malloc(sizeof(float), alignment));
+  samples_total_.reset((float *)volk_malloc(sizeof(float), alignment));
+  power_total_.reset((float *)volk_malloc(sizeof(float), alignment));
   std::vector<std::string> model_server_parts_;
   std::vector<std::string> text_color_parts_;
   boost::split(model_server_parts_, model_server, boost::is_any_of(":"),
@@ -440,16 +440,16 @@ void iq_inference_impl::process_items_(COUNT_T power_in_count,
                samples_since_tag_ += batch_, sample_clock_ += batch_,
                consumed += n_vlen_) {
     COUNT_T j = (power_read + i) % sample_buffer_;
-    volk_32f_index_max_16u(max_.get(), power_in, batch_);
-    float power_max = power_in[*max_];
-    if (power_max < min_peak_points_) {
+    // Gate on average power.
+    volk_32f_accumulator_s32f(power_total_.get(), power_in, batch_);
+    if (*power_total_ / batch_ < min_peak_points_) {
       continue;
     }
     // We might get all zero samples if squelched externally - though
     // we will receive non zero power values.
     const float *in_floats = (const float *)&samples_lookback_[j * batch_];
-    volk_32f_accumulator_s32f(total_.get(), in_floats, batch_ * 2);
-    if (*total_ == 0) {
+    volk_32f_accumulator_s32f(samples_total_.get(), in_floats, batch_ * 2);
+    if (*samples_total_ == 0) {
       continue;
     }
     if (n_inference_ > 0 && ++inference_count_ % n_inference_) {

lib/iq_inference_impl.h

@@ -236,8 +236,8 @@ typedef struct output_item {
 class iq_inference_impl : public iq_inference, base_impl {
 private:
   boost::scoped_array<gr_complex> samples_lookback_;
-  boost::scoped_ptr<float> total_;
-  boost::scoped_ptr<uint16_t> max_;
+  boost::scoped_ptr<float> samples_total_;
+  boost::scoped_ptr<float> power_total_;
   pmt::pmt_t tag_;
   COUNT_T vlen_;
   COUNT_T n_vlen_;