I Spent $12,000 Learning These Signal Generator Setup Mistakes (So You Don't Have To)

The Setup That Almost Cost Me My Q3 Budget

It was early September 2022. I was sitting in our lab, staring at a brand-new Rohde & Schwarz signal generator. The order had come in the day before—a rush job for a major telecom client, testing the new LTE-Advanced carrier aggregation scheme for their upcoming base station.

I’d been handling orders for seven years. By then, I thought I’d seen every mistake possible. But this project taught me something humbling: you can know the theory perfectly and still get the setup wrong.

Mistake #1: The 'Standard' Configuration Assumption

Here's how it started. The spec sheet said we needed a signal generator capable of 64-QAM and 256-QAM modulation schemes. Our R&S SMW200A could handle it easily. I chose a standard configuration from our internal template—a vector signal generator setup I’d used a hundred times before for simple W-CDMA signals.

I ran the first test. The EVM (Error Vector Magnitude) numbers were… terrible. Like, 4.5% where the client spec demanded < 1.2%. The test engineer shook his head. “This is a no-go,” he said.

I said 'standard LTE configuration.' The signal generator heard 'let's use the default crest factor reduction.' The result: a mismatch so bad we wasted two days re-running the same tests.

We were using the same words ("standard LTE") but meaning drastically different things. Discovered this when the customer's acceptance engineer visited and asked why our test signal didn't match the expected power spectral density.

The fix was simple in hindsight: verify the baseband settings against the standard (3GPP TS 36.141), not against habit. The ATSC 3.0 transmission layer also had a specific pilot pattern requirement we’d missed.

Mistake #2: The Mixer and Modulation Bandwidth Trap

That mistake cost me about $3,200 in direct delays and redo time. But the next one was worse.

In November 2022, I had an order for an R&S FSW signal generator test. The customer needed a broadband signal—> 100 MHz bandwidth for 5G NR testing. Our R&S SMW200A's internal baseband generator can do 2 GHz RF bandwidth, but the I/Q modulation bandwidth is limited.

I thought: "It's a premium generator, it'll handle it." The generator's I/Q modulator said: "I can process 200 MHz of modulation bandwidth, but your in-phase and quadrature data is missing a critical filter stage."

The output? Completely distorted. The ACPR (adjacent channel power ratio) was -38 dBc. Acceptable is below -45 dBc for this standard.

So glad I caught this before the client test. Almost shipped the report with the spurious emissions data, which would have portrayed our Rohde & Schwarz test setup as unreliable. (note to self: never trust default filter settings on complex multi-channel signals).

The correction involved downloading the proper 5G NR test firmware (FSW-K144 option) and manually specifying the modulation bandwidths and filter type (for instance, a Root Raised Cosine filter with a roll-off factor of 0.2).

The Big One: The $12,000 Vendor-Shipping-Blame Game

The most frustrating part of this entire saga came in Q1 2024. We'd started a production-run test for a defense contractor, requiring a very specific multi-tone signal for a radar application. The configuration was straightforward: a 16-tone multi-carrier signal on my R&S SMW200A, with precise equal amplitude.

I set it up, ran the test, and the radar's phase noise measurements were off by 8 dB. Eight. The vendor (our contract test house) immediately blamed our signal generator's phase noise performance. “Your Rohde & Schwarz generator must be faulty,” they said. “We can't get a clean enough signal.”

After the third rejection from them, I was ready to give up on the entire multi-tone approach. What ultimately helped was a deep dive into the generator's internal instrument setup documentation.

The issue? The Multitone application on the SMW200A had a hidden setting—a “Signal” → “Optimize for” → “PEP/CCDF for OFDM” mode. Unbeknownst to me, this default optimization was shifting the peak-to-average power ratio (PAPR) away from the equal-tone ideal we needed for the radar test. The resulting noise floor was masking the legitimate phase noise of the DUT (device under test).

Setting it to “Optimize for single-carrier” immediately dropped the noise floor by 7 dB. (Dodged a bullet: I was one click away from authorizing a $15,000 replacement generator purchase.)

1. The Generator: R&S SMW200A vector signal generator
2. The Error: Default multitone optimization creating a 7 dB noise floor penalty
3. The Cost: $8,000 in vendor re-test fees + 2 weeks project delay
4. The Lesson: Never accept default settings when a specific measurement type (here, phase noise) is the critical spec.

Post-Mortem: The Checklist That Saved My Reputation

After the third rejection in Q1 2024, I created our pre-check list. The most critical rules now:

1. Verify vs Standard, not Memory: Before any multi-tone, OFDM, or custom waveform test, I physically print the relevant section of the 3GPP or IEEE standard (sometimes just the key parameters). Do I need an ACLR margin of 5 dB? Is that in the spec?
2. Check Your Signal's 'CV' vs 'EV' Filters: A major source of error. The R&S generators often apply a channel filter (for EVM) and a measurement filter (for spectral flatness). They are not the same. If you use the wrong one, the numbers will be wrong.
3. Always, always, always run a calibration loop: Use the internal calibration signal (the generator’s self-test) before running the client’s signal. I wasted $1,200 last month on a test that failed because I hadn't run the I/Q calibration that morning. The temperature drift from overnight had shifted the I/Q offset.

We've caught 47 potential errors using this checklist in the past 18 months. Not a single one was a 'stupid' mistake—they were all configuration traps that looked right but weren't aligned with the standard's specific measurement points.

Lessons For Anyone Using High-End Test Gear

If you’re using a Rohde & Schwarz signal generator—be it a classic SMW200A, an SGS100A, or a modern R&S SMM100A—don’t make my mistake. The hardware is phenomenal. The I/Q modulator is superb. The problem is always in the software settings and configuration.

When the specs say "vector signal generator capable of 5G NR," that’s true. But the manufacturer assumes you know to load the correct application firmware and filter profiles.

When I switched from the default 'Universal' signal generator profile to the dedicated '5G NR' profile (and manually verified the CCDF optimization), client test acceptance rates improved by roughly 25%. The $12,000+ I spent on those mistakes in 2022-2024 feels like an expensive tuition. I’m sharing this so you skip the tuition and just take the license test.

(Prices as of mid-2024; verify current rates with your R&S representative.)