It’s time to shop for a new oscilloscope. My trusty HP1631d has served me well, but it’s almost 30 years old, and it’s not even really an oscilloscope but a logic analyzer. This boat anchor is big, heavy, loud, cumbersome, and feature-starved. I’ll probably still keep it around for the logic analyzer functions, but for traditional oscilloscope work I’d like something more modern.
Fortunately, this is a great time to be a hobbyist looking for a low-cost scope. There’s practically a tidal wave of entry-level digital scopes coming out of China today, with plenty of good options in the $300-$400 range. The only problem is there are so many options, it can be hard to sift through the blizzard of information and specs in order to choose one!
Why a new entry-level scope, instead of a used digital or analog scope from somebody like Tektronix, which might be purchased for the same price or less? There are some great values on eBay and Craigslist for sure, and for someone on a minimal budget, I think an older used scope is the best option for getting started in electronics. But the major drawback of all those used scopes is that they’re invariably big, heavy, loud boat anchors. They also lack modern connectivity options like USB. I’ve already got a boat anchor, thank you.
When shopping for a scope, there are three primary specs that get the most attention: bandwidth, sample rate, and memory depth. I had only a hazy understanding of how these were related until I began studying a few days ago.
Bandwidth
Bandwidth determines the fastest signal you can measure with the scope. It’s determined by the scope’s front-end electronics, as well as by the probes you’re using. But does that mean a 50 MHz scope is good for measuring signals up to 50 MHz, and useless for anything beyond that? No. The definition of bandwidth for oscilloscopes is actually very specific: it’s the frequency at which the measured amplitude of the input signal is decreased by -3dB (about 30%) of its original value. If you view a 50 MHz sine wave that’s 1v peak-to-peak on a 50 MHz scope, the scope will show 0.7v peak-to-peak. You can view signals faster than 50 MHz on the scope, but they’ll be attenuated even more than -3db. Signals slower than 50 MHz will be attenuated too, but by less than -3db. There’s no magic cutoff at 50 MHz where the scope just stops working.
Sometimes you’ll also see a scope’s rise time advertised. This is the time the scope requires to move from 10% to 90% of a new voltage input. By doing some math, you can demonstrate that bandwidth in GHz is approximately 0.35 / rise time in nanoseconds. So a scope with a 7 ns rise time should have a bandwidth of 0.35 / 7 = 0.05 GHz = 50 MHz.
OK, a little attenuation doesn’t sound so bad. But typically you won’t be measuring sine waves, you’ll be measuring things like square waves. You probably know that square waves are composed of an infinite series of sine waves at odd harmonics of the fundamental frequency. A 40 MHz square wave can be thought of as a 40 MHz sine wave, plus 1/3 of a 120 MHz sine wave, plus 1/5 of a 200 MHz sine wave, plus 1/7 of a 280 MHz sine wave, etc.
If you view that 40 MHz square wave on a 50 MHz scope, it can see the 40 MHz fundamental sine wave, but all the harmonics are above its bandwidth, and are attenuated into insignificance. The result is that the 40 MHz square wave looks like a 40 MHz sine wave when viewed on the 50 MHz scope. That’s really bad. If all you care about is knowing the signal’s frequency, then I guess it’s OK, but the actual shape of the signal is completely lost. If you want to measure the signal’s rise time or duty cycle, you just can’t do it on that scope.
This explains why you need a scope whose bandwidth isn’t just strictly higher than the frequency of the fastest signal you’ll be measuring, but much higher. How much? A common rule of thumb is 4x-5x higher. At 5x, the first two harmonics of the square wave will fall within the scope’s bandwidth, so that seems like a good place to draw the line. But really I think this is one of those “it depends” situations. If almost all the work you do is at 10 MHz or below, but you occasionally have some elements that run as fast as 30 MHz, then you’ll probably be fine with a 50 MHz scope. It just means that when you do need to examine a 30 MHz signal, it will be as if looking through a clouded glass, where the true shape and amplitude of the signal are partly obscured. If you just need to know whether there’s any signal at all, or what its frequency is, then that’s fine. But if you need to look for small glitches or timing problems at 30 MHz, it may not be.
Sampling Rate
All modern oscilloscopes are digital, and sample the input signals millions or billions of times per second. But how fast does the sampling rate need to be, and what’s its relationship to the bandwidth? When I first began looking into it, I thought sampling rate only needed to be 2x bandwidth (Nyquist). While that’s enough to reconstruct the input signal’s frequency, it won’t tell you anything about its shape. At 2x, you’ll get one sample during the high part of each period, and one sample during the low, but you’ll have no idea if the signal is a sine or a square or a lumpy shape. The common rule of thumb is that the sampling rate should be 4x-5x the bandwidth. So if you typically need to examine 20 MHz signals, you should look for a scope with 100 MHz bandwidth and 500M samples/sec.
On most scopes, the sampling rate is cut in half when using two channels instead of one.
Memory
All those digital samples have to be stored somewhere. With more memory, the scope can store more of them. The obvious result of having more memory is that the scope can capture a longer time duration in a single acquisition, which can be useful if you need to debug something like a lengthy serial communication. Or it can store more samples during the same time duration, making it possible to zoom in on the acquired waveform to examine small details.
Not all oscilloscope memory is created equal. Many scopes have standard memory and “long memory”, which operate at different speeds. The long memory stores more samples, but can’t store them as fast as standard memory, so the sampling rate must be reduced when long memory is used. Other scopes have long memory that operates at full speed. This is an important consideration that’s generally not mentioned anywhere in the scope’s advertised specs, but must be inferred from careful reading of the manual or datasheet.
Bandwidth + Sampling Rate + Memory = Confusion
It’s important to recognize that a scope’s advertised sampling rate is its maximum sampling rate, but the actual sampling rate may be much lower. As described above, the use of two channels or slower “long memory” are some reasons that the sampling rate may be reduced below the maximum. But the main reason is that the rate must be reduced when the capture duration increases, in order to keep the total number of sample points small enough to fit in memory. For example, if you want a 50 microsecond per division timescale, that’s a 500 microsecond capture duration. If your scope has 10K points memory, you can only sample the signal at 10000 / 500 us = 20M samples/sec, regardless of how fast the scope’s maximum sample rate is.
When the scope’s sample rate is reduced, its effective bandwidth is also reduced. At 20M samples/sec, the scope has no hope of capturing signals faster than 10 MHz, and won’t be able to faithfully capture the shape of signals faster than about 2 or 2.5 MHz. This leads to the conclusion that memory size constrains bandwidth. That may be old news to some, or a total surprise, as it was for me. With a small memory, at longer timescales, a fancy 200 MHz scope may essentially turn into a 10 MHz scope.
Examples in the Real World
Take a look at this table, which I found on an oscilloscope forum. It compares two popular entry-level oscilloscopes with 100 MHz bandwidth and 1Gs/sec sample rate: the Rigol DS1102e, and the Owon SDS7102. The specs are the same, so you might expect they would have the same basic capabilities for signal capture, but that’s not true. The Owon has 10M points memory per channel, while the Rigol has 1M points shared between both channels. The Owon memory also runs at full speed, while the Rigol’s long memory runs at half speed. The red line shows the point as which the effective bandwidth is reduced below 100 MHz.
The result of 20x more memory and 2x faster memory is a big performance advantage for the Owon. In dual channel mode using the full memory capacity, the Owon can do 500Ms/sec while the Rigol can only do 250Ms/sec. At longer timescales, the difference grows larger. At 1ms/division, the Owon can still do 500Ms/sec while the Rigol is slowed to only 20Ms/sec – a 25x difference! You would never know this if you only looked at the advertised bandwidth and sample rate figures from the manufacturers.
Does this mean the Owon is a 25x better scope than the Rigol? Definitely not. For one thing, most reviewers seem to feel the Owon’s UI design is somewhat clunky and confusing compared to the Rigol. But even looking just at measurement capabilities, you could definitely argue that capturing at 1ms/division while retaining 100MHz bandwidth isn’t very useful in the real world. Most of the time, if you want to examine high speed signals in the greatest detail, you’ll capture them with a short timescale like 50ns, where the sample rate difference between the two scopes is only 2x. And if you’re just looking at a single edge or small number of signal periods, then you won’t need the Rigol’s long memory, and there won’t be any sample rate difference at all.
Bandwidth Hacking
One wrinkle that complicates the shopping process is that some scopes from Rigol and Hantek can be “hacked” in software to unlock higher bandwidths. I first read about these hacks a couple of years ago, but thought they were bullshit. Sure, you might unlock the menu options for higher bandwidths, but if the scope’s ADC and other front-end components only had 60 MHz bandwidth, there’s no way a software change could magically make it into a 200 MHz scope. But after doing much more research, it seems that all scopes in Rigol’s DS1000 line and Hantek’s DSO5000 line are actually 200 MHz scopes, with some models crippled in firmware to limit their bandwidth. People smarter than me have done bandwidth measurements on the hacked scopes, and even disassembled 200 MHz and 60 MHz scopes from the same family. They proved that the internal electronics are the same, and a 60 MHz scope hacked to 200 MHz really does have 200 MHz bandwidth.
Assuming such hacks really work, are they ethical? Are they like pirating software (which I don’t), or more like unlocking your CPU multiplier so you can overclock it (which I do)? Or like jailbreaking your iPhone? My moral compass doesn’t have any problem with it, but I’ll reserve the right to change my mind.
The Contenders
After staring at various oscilloscopes until I was blurry-eyed, I’ve narrowed the list of contenders to just a few.
Rigol DS1052e / DS1102e – This scope has been around for several years now, and has a pretty good reputation. The 50 MHz version can be had for about $300, and can be hacked to 100 MHz. I even used one of these at work for a while. But the DS1000 line is getting a bit old and crufty, with a smaller screen and fewer features than some of the newer competition. I’ve more-or-less ruled it out.
Owon SDS7102 – The Owon has the largest, fastest memory of any entry-level scope today, with 10M points per channel full-speed memory. It also has a very nice 800 x 600 8″ display. It can be purchased for about $420 including shipping, and reviewers on Amazon.com give it an average rating of 4.5 stars. I was very excited about this scope initially, but reviews on electronics web sites haven’t been as enthusiastic as Amazon’s. Many people complain about the user controls being somewhat clunky and confusing. It sounds like the interface isn’t terrible, just kind of annoying. Is it worth being annoyed by your scope every time you use it, in exchange for better sample rates or longer captures in the rare cases when you need them? Probably not.
Hantek DSO5102B – This same 100 MHz scope is also sold under several other brand names and model numbers. There’s an epic 2000 post discussion thread about this scope on the EEVblog forums, describing how to hack it to 200 MHz, modify the interface screens, and tweak other features. Even unmodified, it seems to be a fairly popular and easy to use scope, with most people scoring it best among the entry-level scopes for user interface and features. It’s about $370 including shipping. This is my leading choice at the moment. I was almost ready to buy it before I watched Dave Jones’ video review of it on EEVBlog, where he initially seemed quite pleased with it before he ran into some problems, and finally gave it a thumbs down.
Hantek DSO5062B – This is the 60 MHz version of the Hantek line, and sells for about $345 shipped – $25 less than the 100 MHz model. Both can be hacked to 200 MHz. 100 MHz is really plenty for me, so is it worth $25 to avoid the hassle of hacking the 60 MHz version, and any possible problems it might cause? Yeah, I think it is, so the DSO5062B is probably out of the running.
Hantek DSO5102P – This is the 100 MHz model, but without long memory. It’s $320 shipped, or $50 less. How useful is that long memory, really? There’s no question it could be useful in some situations, but most of the time I’m guessing I wouldn’t use the long memory at all. Why pay for a feature you’re not going to use? Saving $50 sounds good, but when I run into a situation where the long memory would have been useful, I’ll probably wish I hadn’t been so cheap.
Hantek DSO5072P – This scope is 70 MHz, without long memory, but otherwise the same as the other Hanteks. It too can be hacked to 200 MHz, and it’s only $270 shipped. That’s $100 cheaper than the DSO5102B. For $100, would I go through the headache of hacking the scope, and give up the long memory? Hmmm, maybe.
Rigol DS2072 – This last scope is really in a different class than all the others. The major specs are about the same, but the build quality, features, and overall polish are much higher than the others. Dave Jones made an EEVBlog review where he positively gushed about this scope. The only problem is that it’s $840, and that’s for the 70 MHz version! The 100 MHz version is a smooth $1143. But like the other Rigols, the 70 MHz model can be hacked to 200 MHz.
What do you get for $840 vs $350 or so for the other scopes? For starters, the sample rate is 2Gs/sec, as opposed to 1Gs/sec on all the other scopes I mentioned. Long memory is 14 million points, expandable to 56 million. Jitter, noise floor, and all those other details are much better than with the other scopes. It’s a so-called DPO (digital phosphor oscilloscope), where many copies of a periodic waveform are overlaid onscreen with an intensity gradient that’s similar to an analog scope display. Its waveform refresh rate is extremely high, at around 40,000 waves/second, making it great for catching rare glitches. It always runs the sampler at the maximum sample rate, doing averaging on-the-fly when necessary to keep the number of stored samples small enough to fit in memory. It has some very, very cool waveform analysis features. And it even has some basic logic analyzer functions, like automatic decoding of SPI and I2C traffic.
If Santa Claus were bringing me a free oscilloscope, I’d ask for the Rigol DS2072. But if I’m paying out of my own pocket, I’m not convinced all those nice features justify a 2.5x price difference. At the end of the day, I’ll still be looking at the same basic signal on either scope. Bells and whistles are great, but maybe not 2.5x great.
Your Toyota vs. my Horse
Shopping for a new entry-level scope to replace my HP1631d has been funny at times. Even the worst of the contenders are still miles ahead of the 1631d in most respects. It’s amusing to read complaints about scopes that “only” refresh 40 waveforms/sec, when my current scope can barely manage 2 waveforms/sec. Or to read discussions about whether the Hantek –P models are hamstrung by having “only” 40K points of sample memory, when my current scope has just 1K points. 1K!
