[ratings]
In GMSK modulation (used in GSM and DECT standard), a GMSK signal is generated by shaping the information bits in NRZ format through a Gaussian Filter. The filtered pulses are then frequency modulated to yield the GMSK signal. GMSK modulation is quite insensitive to non-linearities of power amplifier and is robust to fading effects. But it has a moderate spectral efficiency.
An expression for theĀ Gaussian Filter with 3dB Bandwidth is derived here.
The requirements for a gaussian filter usedĀ for GMSK modulation inĀ GSM/DECT standardĀ Ā are as follows,
$latex T = \text{ bit duration}&s=2&fg=0000A0$
$latex B =3\text{ dB Bandwidth of the filter}&s=2&fg=0000A0$
$latex BT =0.3 \text{ for GSM}&s=2&fg=0000A0$
$latex BT =0.5 \text{ for DECT}&s=2&fg=0000A0$
Now the challenge is to design a Gaussian Filter fG(t)Ā that satifies the 3dB bandwidth requirement i.e. in the frequency domain at some frequency f=B, the filter should posses -3dB gain ( in otherwords =>Ā half power point locatedĀ at f=B)
The probability density function for aĀ Gaussian Distribution with mean=0Ā andĀ standard deviation=ĻĀ Ā is given by
$latex f(t) = \frac{1}{\sqrt{2 \pi \sigma}} e^{ -\frac{t^{2}}{2 \sigma^{2}} }&s=2&fg=0000A0$
The expression for the required Gaussian Filter can be obtained by choosing the variance of the above mentioned distribution so that the Fourier Transform of the above mentioned expression has a -3dB power gain at f=B.
The fourier transform of the above mentioned expression is
$latex F[f(t)]=e^{-2 \sigma^{2} ( \pi f) }&s=2&fg=0000A0$
Setting f=B,
$latex e^{-2 \sigma^{2} ( \pi B) } = \frac{1}{\sqrt{2}} \Rightarrow \sigma = \frac{\sqrt{ln 2}}{2 \pi B}&s=2&fg=0000A0$
$latex \therefore f_{G}(t) = \sqrt{ \frac{2 \pi }{ ln 2}} B e^{ – \frac{2}{ln2}(\pi B t)^{2}} &s=2&fg=0000A0$
See also :
[1] Correlative Coding – Modified Duobinary Signaling
[2] Correlative Coding – Duobinary signaling
[3] Nyquist and Shannon Theorem
[4] Correlative coding ā Duobinary Signaling
[5] Square Root Raised Cosine Filter (Matched/split filter implementation)
[6] Introduction to Inter Symbol Interference
External Resources:
[1] The care and feeding of digital, pulse-shaping filters – By Ken Gentile
[2] Inter Symbol Interference and Root Raised Cosine Filtering – Complex2real
I think you’re missing a square on (pi*f) in the Fourier transform and on the sigma under the square root in the PDF.
f(t) should be 1/sqrt(2pi) * 1/š -not- 1/(sqrt(2pi*š)
The Fourier transform result should have the (pi*f) term squared as Chris stated
Wolfram-alpha yields the F[f(t)] as:
e^-½(2Ļšf)² = e^(-2ϲšĀ²f²)
Yes, you are right Chris.