diff --git a/docs/user_guide/quantization_usage.rst b/docs/user_guide/quantization_usage.rst
index 2b2f150991c4450ffb92caaa6a9953d1a9b8b90c..0e4246db8a43939b24d8c382a2afa3aa6e48bbfd 100644
--- a/docs/user_guide/quantization_usage.rst
+++ b/docs/user_guide/quantization_usage.rst
@@ -29,29 +29,42 @@ MACE supports post-training quantization if you want to take a chance to quantiz
 This method requires developer to calculate tensor range of each activation layer statistically using sample inputs.
 MACE provides tools to do statistics with following steps:
 
-	1. Convert original model to run on CPU host without obfuscation (by setting `target_abis` to `host`, `runtime` to `cpu`, and `obfuscate` to `0`, appending `:0` to `output_tensors` if missing in yaml config).
-	E.g.,
+	1. Convert original model to run on CPU host without obfuscation (by setting `target_abis` to `host`, `runtime` to `cpu`,
+	and `obfuscate` to `0`, appending `:0` to `output_tensors` if missing in yaml config). E.g.,
 
 	.. code:: sh
 
 		python tools/converter.py convert --config ../mace-models/inception-v3/inception-v3.yml
 
 
-	2. Log tensor range of each activation layer by inferring several samples.
+	2. Log tensor range of each activation layer by inferring several samples on CPU host. Sample inputs should be
+	representative to calculate the ranges of each layer properly.
 
 	.. code:: sh
 
-		python tools/converter.py run --config ../mace-models/inception-v3/inception-v3.yml --example --quantize_stat --input_dir samples > range_log
+		# convert images to input tensors for MACE
+		python tools/image/image_to_tensor.py --input /path/to/directory/of/input/images
+			--output_dir /path/to/directory/of/input/tensors --image_shape=299,299,3
 
+		# rename input tensors to start with input tensor name
+		rename 's/^/input/' *
 
-	3. Calculate overall range of each activation layer by specifying percentage cutoff.
+		# run with input tensors
+		python tools/converter.py run --config ../mace-models/inception-v3/inception-v3.yml --example
+			--quantize_stat --input_dir /path/to/directory/of/input/tensors > range_log
+
+
+	3. Calculate overall range of each activation layer. You may specify `--percentile` or `--enhance` and `--enhance_ratio`
+	to try different ranges and see which is better. Experimentation shows that the default `percentile` and `enhance_ratio`
+	works fine for several common models.
 
 	.. code:: sh
 
-		python mace/python/tools/quantization/quantize_stat.py --log_file range_log --percentile 5 > overall_range
+		python mace/python/tools/quantization/quantize_stat.py --log_file range_log > overall_range
 
 
-	4. Convert quantized model (by setting `quantize` to `1` and `quantize_range_file` to the overall_range file path in yaml config).
+	4. Convert quantized model (by setting `target_abis` to the final target abis, e.g., `armeabi-v7a`,
+	`quantize` to `1` and `quantize_range_file` to the overall_range file path in yaml config).
 
 
 .. note::
diff --git a/third_party/nnlib/libhexagon_controller.so b/third_party/nnlib/libhexagon_controller.so
index ee742a36cd466845e97d50871ae749bf1ec97769..a5d25ee49ada8f23ce5314ce7e5dea9ac361036e 100755
Binary files a/third_party/nnlib/libhexagon_controller.so and b/third_party/nnlib/libhexagon_controller.so differ
diff --git a/third_party/nnlib/v60/libhexagon_nn_skel.so b/third_party/nnlib/v60/libhexagon_nn_skel.so
index 96645ccae17161cdb5aa4efbe5fa814cb4012c8b..b8e0c3b1685420df46a93b115cbba947ddba8b6d 100755
Binary files a/third_party/nnlib/v60/libhexagon_nn_skel.so and b/third_party/nnlib/v60/libhexagon_nn_skel.so differ
diff --git a/third_party/nnlib/v66/libhexagon_nn_skel.so b/third_party/nnlib/v66/libhexagon_nn_skel.so
index 4f86b5a87519d674372b8cf23150475a29969214..054455a8b81eaa2ef72311428727b29e230751f0 100755
Binary files a/third_party/nnlib/v66/libhexagon_nn_skel.so and b/third_party/nnlib/v66/libhexagon_nn_skel.so differ