Preserving dynamic range in fixed-point representation in 5G
Hreitani, Husam (2020-05-25)
Preserving dynamic range in fixed-point representation in 5G
Hreitani, Husam
(25.05.2020)
Julkaisu on tekijänoikeussäännösten alainen. Teosta voi lukea ja tulostaa henkilökohtaista käyttöä varten. Käyttö kaupallisiin tarkoituksiin on kielletty.
avoin
Julkaisun pysyvä osoite on:
https://urn.fi/URN:NBN:fi-fe2020100277789
https://urn.fi/URN:NBN:fi-fe2020100277789
Tiivistelmä
This thesis proposes a method to preserve the dynamic range of frequency samples power
in the physical layer of 5G base-station by using an exponent. Keeping the resource
usage to its minimum while increasing performance and preserve a high Signal-to-Noise
ratio SNR. Operations such as addition and subtraction on samples with different
exponent is made possible by implementing a common exponent block that unifies
exponent with varying powers. The scope also examines multiple rounding mechanisms
to keep quantization as low as possible. Unlike Block Floating-Point technique which
usually operates on the multiplication stages of fast Fourier transform, Dynamic scale
gives more flexibility by separating block floating-point into two parts, the first part
operates on a sample level extracting individual exponents, while the second extracts the
maximum exponent to execute an operations involving more than one sample. A test
case implementation is proposed to give a performance comparison against full precision
floating point model in Matlab.
in the physical layer of 5G base-station by using an exponent. Keeping the resource
usage to its minimum while increasing performance and preserve a high Signal-to-Noise
ratio SNR. Operations such as addition and subtraction on samples with different
exponent is made possible by implementing a common exponent block that unifies
exponent with varying powers. The scope also examines multiple rounding mechanisms
to keep quantization as low as possible. Unlike Block Floating-Point technique which
usually operates on the multiplication stages of fast Fourier transform, Dynamic scale
gives more flexibility by separating block floating-point into two parts, the first part
operates on a sample level extracting individual exponents, while the second extracts the
maximum exponent to execute an operations involving more than one sample. A test
case implementation is proposed to give a performance comparison against full precision
floating point model in Matlab.