Hello,

When using an IDEAL VCO (without noise) in VerilogA to generate a frequency modulated signal, then the instantaneous frequency of that signal is proportional to the modulating input signal.
I can use VerilogA to measure the period of the FM signal by using the @cross function for each zero crossing. Thus I can get the instantaneous frequency. I then export these values at my sampling rate Fs to Matlab where I do an FFT.

My question is regarding the noise floor. The maximum SNR I can get is somewhat in the range of 90dB for a given VCO gain and input amplitude. Why only so little? What determines the noise floor as everything is ideal and I am not introducing any quantization either?
With errors I mean that the VCO is ideal, the period determination of the FM signal is ideal and my Fs is at much greater than the free running frequency of the VCO (just making sure I am exporting enough values to matlab).
What determines the noise floor?

Thanks for your thoughts.

Thanks for the response ACWWong.

ACWWong wrote on May 8^th, 2007, 6:47am:

I am not certain I understand how to do this. How do you make sure the simulated data points from Spectre coincide with the rising edge of the sampling clock? I assume you referring to  $bound_step = $bound_step(1/Fs); ?

ACWWong wrote on May 8^th, 2007, 6:47am:

If you mean setting the "maxstep" option then I have already done so. For example, my Fs is approx. 200MHz and the "maxstep" I tried all the way down to 1e-10 without improvement in the noise floor.

Hmmm, I tried setting the maxstep parameter down to 50e-12, errpreset=conservative, $bound_step(0.1*period); without any improvement in the noise floor. I also used the maximum tolerance option for the @cross function. (see code below)

Code:

simulator lang=spectre
ahdl_include "widthcounter.va"
ahdl_include "fmeasure.va"

parameters FS=200e6

Vin  (in  0)  vsource type=sine ampl=0.5 freq=8.4e3 dc=0 sinedc=0
Vclk  (clk 0)  vsource type=pulse period=1/FS rise=1p fall=1p
sdvco (out in clk) sdm Kvco=0.5e6 f0=4*4.352e6 period=1/FS

resp tran stop=8/8.4e3 skipdc=yes errpreset=conservative maxstep=50e-12
 

Code:

`include "disciplines.vams"
`include "constants.vams"

module sdm (out, in, clk);

input in, clk; voltage in, clk;
output out; voltage out;

parameter real f0=4*4.352e6;		// Carrier frequency VCO
parameter real Kvco=0.5e6;		// Gain VCO
parameter real period=1/200e6;		// Sampling frequency

integer fptr1;
real TFM, last_timee, Te;
real freq, phase, last_time, current_time;

analog begin

	@(initial_step)
	begin
		fptr1=$fopen("...path to file");
	end

	$bound_step(0.1*period);
	freq = f0+Kvco*V(in);
	  phase = 2*`M_PI*idtmod(freq, 0.0, 1.0, -0.5);

	@(cross(phase + `M_PI/2, +1, 1e-12, 1e-12) or cross(phase - `M_PI/2, +1, 1e-12, 1e-12))
		begin
			current_time = $abstime;
			if (last_time > 0.0) TFM = period/(current_time - last_time);
			last_time = current_time;
		end

	@(cross(V(clk) - 0.5, 1)) begin
		$fstrobe(fptr1, "%g", TFM);
	end

	@(final_step) $fclose(fptr1);

	V(out) <+ phase;

end
endmodule
 

Do you know what else I might try?
Cheers

Ken Kundert wrote on May 9^th, 2007, 1:34pm:


Ken, when looking at the difference in time that I get when using @cross and last_crossing it was about 30f.
So subtracting this difference as shown below gave me a some improvement (noise floor drops by about 10dB).

Code:

............
............
bla=last_crossing(phase,0);

@(cross(phase, 0, 1e-12, 1e-12)
		begin
			current_time = $abstime;
			diff = current_time - bla ;
			TFM = period/(current_time - last_time - diff);
			last_time = bla;
		end
...........
...........
 

Is this what you had in mind?
Thanks,

Ps. Why would the last_crossing function give more accurate results than the @cross function, i.e what happens in the simulator settings?