Fractional Divider in verilog

Fractional Divider

This design note shows the derivation of the equation for a fractional divider (FD) and provides the verilog code implementation. This code has been tested to work on a Spartan 3AN and is available on Bitbucket.

A fractional divider gives higher resolution in a digital PLL without increasing the clock frequency. As example, a common divider will have an output frequency of fout= fosc/N. From this, the small signal gain is df/dN = -fosc/N^2 = -fref/N. This term is the frequency resolution in Hz per count or also known as the channel spacing.

Derivation

The TI Technical Brief SWRA029 shows a derivation. The basic principle is to count N+1 counts for m cycles and N counts for M-m cycles. The resulting total number of counts is the sum {(N+1)*m + N*(M-m)} and the total cycles are m + (M-m) = M. From this simple equation the average counts per cycle is Nbar = N + m/M.

The output frequency is fout = fosc/Nbar = fosc/(N +m/M) = (fosc/N)*1/(1 + m/(N*M)). Observe that the term m/(N*M) is much smaller than 1 so this can be approximated as fref = (fosc/N)*(1 - m/(N*M)). Remember 1/(1+x) ~= (1-x) for x<< 1.

The VCO small signal gain is defined as df/dm = -(fosc/N)/(N*M) = -fref/(N*M). Note this has increased the resolution by a factor of M. This is important to reduce the limit cycle effects of a closed loop system with dead-band, which occurs in a discrete system such as this.

Implementation

M is parameterized in the verilog code by the constant fsze. Using fsze= 3 (bits) yields M=8. In effect we are using the input divided value as a integer with fractional bits, yyyy.xxx.

There are two counters in the verilog implementation. The main counter is dividing by the integer count N. The second counter is down counting from M-1 and rolling over at 0. When the main counter down counts to zero it is loaded with N or N-1 depending on whether m is larger or smaller the the count in the M counter.

This code has been tested with fosc=50MHz, N=68750 and fsze= 3 & 4 in a digital PLL. The code was simulated using Veritakwin and synthesized using Xilinx Project Navigator 14.7.