// // NCO Demo (Numerically controlled Oscillator). // // One type of NCO is based on the idea of a continuously wrapping modulo // counter. The NCO is a programmable modulo counter. For example, if // the MOD input is set to 10, then it'll count 0,1,2,3,4,5,6,7,8,9,0,1,2 etc. // The STEP input is continuously added to the modulo counter. What's important // is not the counter value itself, but the number of times it "wraps". If // you do the math, the counter will wrap at a rate of F*S/M (F is the system // clock frequency, S is the step input, and M is the modulo). Every wrap // occurance is a little pulse. The pulse output could be used as the clock itself // except that it may have a very small duty cycle. Instead, take the WRAP // pulse signal and use it to increment a TICKS counter. The TICKS counter will // then count at the programmed frequency. // // The TICK counter could be used in itself for, say, a TDMA slot counter. Any // number of TICKS LSB bits can be thrown away to "reduce jitter" but also // reduce the rate. This NCO has a little MASK input and an internal OR gate // so you can pick off whichever TICKS bits you wish for your final FOUT output. // // In an open-loop mode, you program your Modulo and Step inputs and you are done. // If you are tracking some other reference (e.g. like in a PLL style circuit) you // would somehow control STEP via some sort of "phase detector" and loop filter. // None of this is shown here. // // Anyway, this is just the guts of the NCO, which can be applied in many, many // ways. // // Oh.. And this NCO seems to be off by a small percentage..?!.. I'm not sure // why. It may be the circuit or the testbench. Let me know if you play with it // and find out. My application is closed-loop, so, I'm not highly motivated // to figure it out. // // Written by tom coonan // module nco ( clk, resetb, step, // Step input is continuously added to the modulo counter mod, // modulo mask, // Mask is ANDed with ticks and gen ORed to produce fout ticks, // Tick counter output fout // Output. ); parameter W_ACCUM = 24; // Width of the Accumulator parameter W_TICK = 8; // Width of the Tick counter. parameter W_STEP = 24; parameter W_MOD = 24; input clk; input resetb; input [W_STEP-1:0] step; input [W_MOD-1:0] mod; input [W_TICK-1:0] mask; output [W_TICK-1:0] ticks; output fout; // Registered outputs // Internals reg [W_ACCUM-1:0] accum, accum_in; reg [W_TICK-1:0] ticks; // *** Modulo Counter *** // Registered outputs reg wrap; wire [W_ACCUM-1:0] sum = accum + step; wire [W_ACCUM-1:0] rem = sum - mod; wire over = (sum >= mod); always @(posedge clk or negedge resetb) if (~resetb) accum <= 0; else accum <= accum_in; always @(over or rem or sum) begin if (over) begin // Wrap! accum_in <= rem; // load remainder instead of sum wrap <= 1; end else begin // No wrap, just add accum_in <= sum; wrap <= 0; end end // *** Tick Counter *** // always @(posedge clk) begin if (~resetb) ticks <= 0; else begin // Whenever Modulo counter wraps, increment the tick counter. if (wrap) ticks <= ticks + 1; end end // *** Masks and final output *** // assign fout = |(ticks & mask); endmodule module ncotest; reg clk; reg resetb; reg [23:0] step; reg [23:0] mod; reg [7:0] mask; wire fout; wire [7:0] ticks; parameter W_ACCUM = 24; // Width of the Accumulator parameter W_TICK = 8; // Width of the Tick counter. parameter W_STEP = 24; parameter W_MOD = 24; nco nco1 ( .clk(clk), .resetb(resetb), .step(step), .mod(mod), .mask(mask), .fout(fout), .ticks(ticks) ); parameter PERIOD_NS = 36; parameter DUMP_ON = 1; real sys_freq; initial begin step = 0; mod = 0; mask = 8'b00000001; // Final Divider for FOUT sys_freq = 1000000000.0/(PERIOD_NS); #300; // Display the basic such as system clock frequency, etc. // $display ("NCO Test. NCO Accumulator width is %0d bits, system clock period is %0d ns (%fMHz).", W_ACCUM, PERIOD_NS, sys_freq ); // Program Modulo and Step for desired frequency. Modulo and Step values should // not be divisible by each other (I'm not mathematically strong enough to // justify this...). Find the ratio of S/M, integerize it, and reduce to lowest // common divisors. Then, multiply up by a big power of 2 so we can get some // resolution on the NCO. // // Let's generate 1Mhz, Fsys*(Step/Mod)/2 = 27777777.777*(S/M)/2 = 1000000 // S/M = 0.072 = 72/1000 = 9/125 (9 * 2^12) / (125 * 2^12) // mod = 125 << 12; // Shift up so we get resolution.. step = 9 << 12; // Shift up so we get resolution.. nco_test (mod, step, 1000000); // Run test for specified interval (units are NS) // Generate 10.24MHz: Fsys*(Step/Mod)/2 = 27777777.777*(S/M)/2 = 10240000 // S/M = 0.73728 = 73728/100000 = 9216/12500 = 4608/6250 = 2304/3125 // mod = 3125 << 10; // Shift up so we get resolution.. step = 2304 << 10; // Shift up so we get resolution.. nco_test (mod, step, 1000000); // Run test for specified interval (units are NS) // Generate 32.768 KHz using TICKS MSB (divide by 8): // Fsys*(Step/Mod) = 27777777.777*(S/M) = 32768*256 // S/M = 0.301989888 =~ 0.302 = 302/1000 = 151/500 // mask = 8'b10000000; // Divide by 256 mod = 500 << 12; step = (151 << 12) - 9566; // Manually Tweaked this to get the right number.. // this is expected since didn't have a good integer // ratio above.. nco_test (mod, step, 1000000); // Run test for specified interval (units are NS) $display ("Done."); $finish; end // Run a single trial of the NCO test. // task nco_test; input [23:0] mod_arg; input [23:0] step_arg; input interval; integer interval; // Use $time.. Make sure timescale is correct! integer start_time; integer fout_edges; begin step = step_arg; // Configure NCO mod = mod_arg; // Count rising edges on FOUT which is the output frequency fout_edges = 0; start_time = $time; // Note our starting time // Loop for at least the specified amount of time while ( ($time - start_time) < interval) begin @(posedge fout); // Wait for an edge on FOUT fout_edges = fout_edges + 1; end // Done. Display results and expected results. $display ("For Mod=%0d(0x%h), Step=%0d(0x%h), Frequency of fout = %f Hz, Expected fout is %f Hz.", mod, mod, step, step, ((fout_edges*1.0)/($time - start_time))*1000000000.0, // measured.. ((step*1.0)/(mod*1.0))*(sys_freq)/(mask*2.0) // expected.. ); end endtask // Let's clock it at about 27 MHz initial begin clk = 0; forever begin #(PERIOD_NS/2) clk = ~clk; end end initial begin resetb = 0; #200 resetb = 1; end initial begin if (DUMP_ON) begin $dumpfile ("nco.vcd"); $dumpvars (0,ncotest); end end endmodule