Department of Electrical
and Computer Engineering,
Wireless Networking and Communications Group,
The University of Texas at Austin,
Austin, TX 78712 USA

karl.nieman@utexas.edu -
bevans@ece.utexas.edu

ICC 2016 Submission on Multi-antenna Baseband LTE Compression (October 16, 2015)

Slides (July 23, 2015)

- design of a low-complexity compression method for LTE and
- validation of this method using an LTE link-level simulation.

Answer #1: Inverse quantization can be implemented as a look up table that maps the quantized codeword index back to floating point.

*Question #2: The variance of the Gaussian distribution is implied
in the choice of thresholds, and it should be estimated or calculated,
how do you estimate or calculate it in the simulation?*o

Answer #2: The variance of the distribution can be measured using a receive signal strength indicator (RSSI) in uplink and is known in the downlink. For downlink using unitary inverse fast Fourier transform and average power constraint across subcarriers, sigma^2 = 1.

*Question #3: There appears to be errors in equation (4).*

Answer #3: Yes, equation (4) has two errors in it:
*N* should have been *L* and
*t _{j}* should have been

*Question #4: Could you give the derivation for equation (7)?*

Answer #4:
The derivation of equation (7)
is available.
The derivation of equation (4) can be performed in the same way.
The key trick here is to replace the denominator in equation (3) with
1/*L* since we have assumed that we have divided the probability density
function into equal probability intervals (for *L* intervals,
they each have 1/*L* probability.

*Question #5: How were to able to control the increase in the word size due to the resampling operation?*

Answer #5: In implementation, performing resampling by using a cascade of an upsampler, filter and downsampler causes an increase in the word size if the filter is a linear time-invariant filter. The increase in word size is due to multiplication and addition operations. Multiplying a five-bit data value by an eight-bit coefficient yields 13 bits without any loss of precision. Instead of using a linear time-invariant filter, we tried median and other rank-order filters which prevents the increase in word size, but at the expense of significant error vector magnitude (EVM) loss. In an implementation, we would recommend using noise-shaping alone, which would give a 3x compression of baseband LTE data with an EVM loss of less than 2%. Please see a more recent presentation on this topic entitled Baseband LTE Compression (July 23, 2015) for a comparison of several baseband LTE compression techniques.

Last Updated 10/16/15.