True Vertical SensitivityThe SDS800X HD has a specified vertical gain range from 500 µs/div up to 10 V/div. Many contemporary DSOs have similar specs, yet only a small minority of them can provide true 500 µV/div as the real highest sensitivity at full resolution. The real sensitivity of many instruments is lower, sometimes significantly so (up to 5 mV/div). As a consequence, anything above the true highest sensitivity is just software zoom and won’t provide full resolution anymore. This might be not that much of a problem for a 12 bit DSO, but 8 bit instruments degraded to 6 bits at 1 mV/div could get problematic. On the other hand, most of those instruments also exhibit high noise levels, so the ENOB (Effective Number of Bits) drops below 6 bits at these higher sensitivities anyway.
For the SDS800X HD, I stumbled across the unexpected property of nearly equal noise levels for all vertical sensitivities from 500 µV/div to 5 mV/div:
SDS824X HD_ND
These numbers are not totally accurate because it proves very difficult to place the markers close to the intended frequency without hitting a minor spur. Consequently, I would think that the noise level is fairly uniform across all the higher sensitivities from 500 µV/div to 5 mV/div and the minima across all measurements would be the best representation of the truth:
Noise 500 µV/div – 5 mV/div :
1 kHz : 232.9 nV/√Hz
3 kHz : 174.2 nV/√Hz
10 kHz : 62.5 nV/√Hz
30 kHz : 16.0 nV/√Hz
100 kHz : 5.7 nV/√Hz
300 kHz : 2.7 nV/√Hz
1 MHz : 2.4 nV/√Hz
10 MHz : 2.5 nV/√Hz
This made me suspicious: does Siglent cheat after all? Are all vertical gain settings below 5 mV/div just fake? First, I’ve checked the raw acquisition data for 500 µV/div vertical gain and found the lowest voltage step to be 1.042 µV.
Time [sec] Value [V] Delta [V]
-4.0000000000E-08 -4,166667E-05 5,208E-6
-3.9500000000E-08 -4,166667E-05 000,000E+0
-3.9000000000E-08 -4,270833E-05
1,042E-6The SDS800X HD has 480 LSB per vertical division (just like the SDS2000X HD), thus 3840 LSB on the visible part of the screen. Since a 12 bit acquisition system provides a total 0f 4096 LSB, there is very little headroom outside the visible screen area.
The interesting part is when we multiply the 1.042 µV resolution with the 480 LSB of one division: 1.042 * 480 ~ 500 µV/div; -> Bingo!
A less accurate, but quicker and simpler method to verify the resolution of the SDS800X HD is using vertical zoom; we can zoom into the noise in dots display mode, thus getting horizontal lines vertically spaced according to the true resolution of the instrument.
At a vertical gain of 500 µV/div and a vertical zoom window at 2 µV/div, we get the following picture:
SDS824X_HD_Resolution_Demo
Since we still have 8 vertical divisions also in the zoom window, the total visible screen height covers 16 µV at 2 µV/div. We can count 15 horizontal lines, hence 16 steps and can conclude that each step has to be close to one microvolt.
Verdict: Siglent don’t cheat. The uniform noise level at and below 5 mV/div is just a property of the integrated PGA (Programmable Gain Amplifier) used in this instrument.
Peak DetectThe peak detection capability of the SDS800X HD is specified as 2 ns. Let’s have a closer look at that.
First a 2 ns wide pulse with 300 mV amplitude and 500 ps rise time in normal acquisition mode at sufficient sample rate (2 GSa/s):
SDS824X_HD_Pulse_W2ns_RT500ps_2GSa_Norm_Zoom
It can be seen that such a narrow pulse is already a bit too much for a 200 (244) MHz oscilloscope; the amplitude has already dropped a bit and pulse width measurement isn’t quite accurate either. As expected, the rise time measurement approaches the scope’s own rise time.
With all these shortcomings, we still get a fairly stable picture – look at the main window and the peak and standard deviations in the measurements statistics.
In the screenshot above, the time base was at 5 ms/div and the sample memory was already at its maximum of 100 Mpts; slowing down the time base any further will inevitably lower the sample rate:
SDS824X_HD_Pulse_W2ns_RT500ps_100MSa_Norm_Zoom
At 100 ms/div and 100 Mpts record length the sample rate has to be decimated to just 100 MSa/s – far too slow for capturing a 2 ns wide pulse. As a consequence, many pulses get lost. In the main window we would expect to see about 1000 pulses at a pulse repetition rate of 1 kHz, but there are actually much less and the amplitudes vary wildly.
This isn’t a very realistic scenario; not many engineers would try to watch 2 ns wide pulses at a time base of 100 ms/div and have to use 2 million times zoom to watch the pulse details. Yet this is where Peak Detect acquisition mode comes into play:
SDS824X_HD_Pulse_W2ns_RT500ps_100MSa_Peak_Zoom
The main window now shows all the pulses; the amplitudes still vary a bit, but at least we don’t miss any pulses anymore. Pulse shape has nothing to do with reality anymore and measurements yield just house numbers. This should be a clear warning to not use Peak Detect for anything serious, as any math and measurements on such waveforms are of artistical value at best.
All that Peak Detect really can do is to hint on any pulses within the record.
Of course peak detection works for even narrower pulses just as well. This is not because the specification is not correct, but the simple fact that a 244 MHz DSO like the SDS824X HD simply cannot process even faster pulses:
SDS824X_HD_Pulse_W1ns_RT500ps_2GSa_Norm_Zoom
This is now a 1 ns wide pulse at maximum sample rate of 2 GSa/s. The amplitude is still 300 mV, yet the SDS824X HD cannot cope with it anymore and the amplitude measurement result has dropped to just 173 mV. The pulse width is still measured as 1.8 ns, so the relative slowness of the frontend widens shorter pulses at the expense of amplitude, hence makes an even faster peak detection unnecessary.
Bode Plot at a glanceInstead of showing an inexpressive first order RC-lowpass filter demonstrating less than 40 dB dynamic in the audio range, I’d rather check the most important characteristics of a Bode Plot: frequency- & dynamic range and accuracy.
For this, I’ve refrained from using inline terminators at the scope inputs but fed them from 50 ohms sources directly via ~25 cm long coaxial cables. The source resistance of 50 ohms, together with the cable and scope input capacitances, forms a first order lowpass filter at ~10 MHz. This can also serve as a warning how even very short cables can introduce significant amplitude errors at relatively low frequencies, as long as a transmission line is not properly terminated.
We can see this characteristic when using the “Vout” mode of the Bode Plot, where we get the absolute amplitude of the DUT output (where the DSO itself represents the DUT).
SDS824X HD_Bode_1M_Vout
The amplitude drops quite significantly above 10 MHz. It is not the 20 dB/decade like a classic first order lowpass – and this is for a number of reasons that I won’t discuss in this article. Bottom line is, that even with very short cables, accuracy of the absolute signal level is gone already at moderate frequencies of a couple MHz.
The phase plot does not resemble this, as it stays within +/-1° up to 120 MHz quite easily. It almost looks like this would not be a minimal phase system, yet it’s just the nature of a multi-channel oscilloscope, where the input signals are always phase aligned.
When using the relative (Vout/Vin) mode (as we usually do), things look completely different:
SDS824X HD_Bode_1M_S21
Bode Plot now shows the difference between reference channel 1 and the other channels. It is indicative of the quality of the SDS800X HD that the differences between channels are really negligible: less than 0.3 dB amplitude error as well as less than 1° phase error up to 120 MHz, and almost no differences between channels 2-4, speaks for itself.
Let’s check the accuracy and dynamic range now. Two signals are used to visualize a 60 dB amplitude difference. This time, 50 ohm inline termination has been used.
SDS824X HD_Bode_50_S21_60dB
There is a significant phase difference, and this comes from the additional 3-stage step attenuator + Inline attenuator + some 50 cm additional coaxial cable for channel 4.
As a final experiment, here is a 100 dB amplitude difference (phase has been adjusted by means of the channel skew parameter):
SDS824X HD_Bode_50_S21_100dB
Noise is getting a major problem, yet amplitude measurements can still yield useable results in the range 100 kHz to ~20 MHz.
The reference level is low (~570 mVrms), hence channel 4 input sees only 5.7 µVrms!
I’ve not nearly exploited the dynamic range of the SDS800X HD, which could handle up to 28 Vrms (but then with beefy external >16 W terminators) if the need should be.
Home
Help
Search
Login
Register
