Accuracy of CROSS() function in Verilog-A

My guess is that this might be something to do with now phase is defined. So can you show the rest of the code (at least the part that is setting the phase variable?).

Also, what is ttol set to?

And what version of spectre are you using?

Regards,

Andrew.

Dear Andrew,

I am using Spectre version :-  sub-version  12.1.0.402.isr5.

The code is a simple one which model a digital Oscillator whose Time period varies with respect to the input voltage V(in).

The code is as below:-

 // VerilogA for VERILOG_A_MODEL, DCO, veriloga

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

module DCO (in, out,out1);

input in;
output out,out1;
electrical in, out,out1,phase_out;
parameter real Vlo=0, Vhi=1.25;

parameter real Vth=0;

parameter real ttol=1p;
parameter real vtol=1p;
parameter real tt = 1p;


parameter real KT=200e6;
parameter real Tv_0= 0.5n;
real n,delay;
parameter real start_delay=1n;

real T_period;
real Time_Span;
real phase,reset;



analog begin

@(initial_step) begin
      V(out) <+ 0;    
      
      delay= start_delay;
 end
 reset =0;
 
 T_period = Tv_0 + KT * V(in) ;
 
 

 V(phase_out) <+ phase;
    
 @(cross(V(phase_out)-T_

I had some trouble running it - I had to change KT to be a smaller number (see the comment in the code below). The key seemed to be moving the definition of phase to above the cross. I think the issue is that it was using the previous iteration's value of phase.

I also changed it to use phase again rather than V(phase_out) (spectre advised me that it had changed it back to a real number for performance reasons anyway).

This was the netlist I used to test it with:

 //

OSC1 (in op1 op2) DCO

v1 (in 0) vsource type=dc dc=1
r1 (op1 0) resistor r=100k
r2 (op2 0) resistor r=100k

ahdl_include "DCO.va"

tran tran stop=1m

 // VerilogA for VERILOG_A_MODEL, DCO, veriloga

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

module DCO (in, out,out1);

input in;
output out,out1;
electrical in, out,out1;
parameter real Vlo=0, Vhi=1.25;

parameter real Vth=0;

parameter real ttol=1p;
parameter real vtol=1p;
parameter real tt = 1p;


// changed KT value because it seemed enormous to me otherwise!
parameter real KT=200e-6;
parameter real Tv_0= 0.5n;
real n,delay;
parameter real start_delay=1n;

real T_period;
real Time_Span;
real phase,reset;



analog begin

@(initial_step) begin
      V(out) <+ 0;    
      
      delay= start_delay;
      reset =0;
 end
 
 T_period = Tv_0 + KT * V(in) ;

// $strobe("T_period is ",T_period);
 
 

 phase= idt(1,0,reset) - delay; // Phase Integration
 reset =0;

// V(phase_out) <+ phase;
    
 @(cross(phase-T_period,+1,ttol,vtol)) begin
       reset=1;
       delay =0;
       $strobe("T_period is ",T_period);
       $strobe("cross val is ",phase-T_period);
       $strobe("EDGE_TIME = %rs \n", $abstime);
      
     end
     
// $strobe("phase is ",phase);


  n= (phase >=0) && (phase < (T_period/2));
     
     V(out) <+ transition(n ? Vhi:Vlo,0,tt);
     V(out1) <+ phase ;
     
     //$bound_step(0.01n);     
     
end
 endmodule

Dear Andrew,

Thanks a lot for your time and reply.

I realised that it is NOT the cross() issue.

The issue lies in idt(). 

When the "phase" crosses "T_period" ( cross() is accurate no doubt ) , "reset" becomes one 1 at that instant.

But idt() becomes zero in the next time point which is NOT controlled one ( it depends upon Spectre simulation).

This leads to period error of the Oscillator.

Same is the case for idtmod(). 

Is there any way I can make idt() = 0 at the exact time instant when "reset" becomes 1 though it is completely abrupt and discontinuous.

Kind Regards,

The conventional way of handling this would be to convert the input voltage to the expected instantaneous frequency, and then integrate that to produce the phase. In this example:

       inst_freq = center_freq + vco_gain * V(vin);
      $bound_step (1.0 / (steps_per_period*inst_freq));

      phase = idtmod(inst_freq,0,1);
      V(vout) <+ amp * sin (2 * `PI * phase);

The phase will ramp up from 0 to 1 and the automatically reset. In this case we're using a sin() to produce a sinusoidal output, but you could also use an @cross or similar to produce a logic output.

Regards,

Andrew.

real iout,vd,temp;

analog initial
state = 0 ;

analog begin
//temp=0;

vd=V(vc,out);
@(cross(vd-3,1))
state=1 ;
@(cross(vd-2,-1))
state=0 ;

$strobe("Vds=%e",vd);

if (state==1)
begin
temp=0.5+0.5*tanh((vd-2)/0.01);
iout = temp*(vd/500)+(1-temp)*(vd/7000);
end
else
begin
temp=0.5+0.5*tanh((vd-3)/0.01);
iout = temp*(vd/500)+(1-temp)*(vd/7000);
end

I(vc,out) <+ iout ;
end

endmodule

I want to implement a hysteresis loop in I_V_characteristics at V=3 and V=4 being transition points. Is there any alternative to cross-function? Please help me out in this 

Why do you need an alternative to the cross function? Surely that does what you want? (the code above appears to have transitions at 3V and 2V).

Andrew.

Actually, I have to implement a hysteresis with the transition happening at 2 and 3 V . The circuit made with this device used should oscillate but this implementation is showing an error. So I thought maybe it is due to cross so I asked for an alternative.