# Easiest way to generate quadrature signal on PCB

I have a chip which needs a quadrature LO. Specifications:

• Frequency range: 1.7 GHz - 2.5 GHz (wider range appreciated)
• Both I/Q differential (i.e. 4 wires in total)
• Each wire 1Vpp
• DC offset of 500mV (capable of driving 50 Ohm, i.e., 10-20mA)
• Matching is 50 Ohm but not perfect (S11 > -10dB)
• Jitter: hundreds of fs rms (like 400fs)
• Phase accuracy: ~1 deg
• If possible: Capability of fast (or even ultra fast) hopping/settling (around 10ns)
• Cost: "does not matter"
• Power: "does not matter"

For my current revision, I use an external signal generator which goes into a 90 Degree hybrid and enters the PCB via 2 SMA connectors (I and Q). Each of these channels go through an on-board 1:1 balun. An RF choke adds DC offset of 500mV.

However, this setup is clunky, the on-board balun (MABA-007871-CT1A40) does not even meet the specs. For my second revision, I would like to generate these signals on-board to simplify the setup.

An RFDAC (like AD9164: 16 Bit, 12 GPBS) would satisfy all requirements but getting this 165-pin BGA beast with its SERDES interface working takes me forever.

So if I relax the fast settling requirement I would assume there are solutions out there. However, I am unable to find any quadrature generators on DigiKey! (I can find I/Q Modulators or Demodulators but I need the generation).

• How much harmonic distortion can you tolerate? I wonder about using a digital clock generator chip and then filtering off the harmonics. Jan 23, 2018 at 1:01
• Must it be a sine wave? Jan 23, 2018 at 1:07
• Two channels of DDS + two RF frequency multipliers. OR Two channels of high bandwidth DDS + up-conversion mixers. Each up-conversion mixer would receive 1.75 GHz CW + one of the DDS's. I'm not an RF engineer, but I think this could work. Jan 23, 2018 at 1:08
• Levels can be sorted out using pads if needed. If true differential signal is needed, you can use a balun on each signal. Jan 23, 2018 at 1:10
• @PeterGreen: It is a clock signal, hence HD is not an issue; the signal is rectified using inverters. However, digital clock generators usually don't satisfy 400fsrms? Also, the main question is about (simple) quadrature output.
– divB
Jan 23, 2018 at 1:57

## Fast flip flops make light work of 0 to 90

Since the target system can accept a non-sine-wave signal, the solution to this problem is actually rather simple -- generate a 3.4 to 5GHz differential clock signal (by hook or by crook, I don't care how you get it), and then use 2 DFFs in the classical quadrature generator configuration, as depicted below (image from this article):

Considering that fast flip flops, such as the 10GHz NB7V52M have differential I/Os, we can elide the inverter (it becomes a differential transposition), yielding the resulting circuit (termination components not shown for clarity's sake):

simulate this circuit – Schematic created using CircuitLab

Parts cost is a non-issue -- I seriously doubt you can get a 5GHz generator for less than the cost of the two flip-flop ICs in the solution.

A wideband analogue phase splitter can be made like this: -

The input stimulus comes from V1 and I and Q outputs are at Va and Vb. Providing you keep the two resistors at a value of $\sqrt{\frac{L}{C}}$ then you get a constant phase shift of exactly 90 degrees between Va and Vb.

So, for the implementation above, the bode diagram is: -

The only down side is that there are amplitude variations in Va and Vb across the frequency range. For instance, at 1.7 GHz, Va is about 2 dB higher than Vb. At 2.5 GHz, Vb is about 1.4 dB bigger than Va.

Just a thought.

Why not using a passive polyphase filter like this one?

Image source: http://article.sapub.org/10.5923.j.msse.20120104.02.html