Derek Song

• DDS generates the waveform by looking up the pre-loaded 2N data in memory. • The reference clock is fixed (fref). The address interval of looking up the memory (Tuning word, “M” in the formule) determines frequency of output (fout).

Profile of DDS(2) • For sinusoidal wave, according to the Sampling Theorem, as long as every period includes more than 2 points (fref ≥ 2fout ), the signal can be reconstructed without distortion, theoretically. • But for square/pulse and arbitrary waveformDDS has its intrinsic defects.

DDS ’ defect(1) • When generating square/pulse, if reference frequency is not exactly the integral multiple of output frequency (i.e. mod(fref /fout )≠0), it will introduce a deterministic jitter equal to one reference clock period. In the figure the reference clock period is 20ns

DDS ’ defect(2) • When Tuning word1 (i.e. fout fref /2N ), some points in lookup table are jumped over. This is inessential for sinusoidal wave, but for arbitrary waveforms with some important detail (e.g. spikes), it may means loss of information.

N Points Lookup Table Tuning word 1 Sampling Rate Conversion • TrueArb outputs data in the N-point lookup table point by point, with the data rate fout *N, then convert the data rate from fout*N to fref. • No missing point, and no DDS intrinsic jitter.

TrueArb(2) • TrueArb shows no one-reference-periodjitter, compared to DDS.

TrueArb(3) • TrueArb shows no missing data, compared to DDS. In the figure the waveform data sequence has 9 spikes each period.

TrueArb(4) • Point by point output, no missing data • Low jitter (<200ps) • Length of data table is variable (8 pts~ 8 Mpts) • Waveform segmentation and sequencing are possible