Best entry level oscilloscope for computer engineering?

[Wuerstchenhund]

Edit: I'm more interested in the architecture then anything else, that's why I initially said I would like to build a 1970s era computer. Computers from this era still had discrete components. Thinking about it more, I think the first step would be to build a simplistic microcontroller from discrete components. This would help me learn things like gates, digital logic, adders, ALUs, microcode, etc. Does anyone know of a training kit that has all of this? I think a lot of this could be done in simulation or on a FPGA today, however, I want to be able to touch it. I learn best through hands on labs.

Sounds good. My recommendation then would be to forget about Raspberry Pis, ATMegas and other modern day complex stuff and start with something simple like an 8051. Yes, it's old but the 8051 is cheap, slow (means you don't need high speed instruments to look at your signals) and simple, and a perfect starting point if you want to learn how computers work. There are various 8051 kits around but you could as well build your own, which is much easier than doing the same with much more complex and faster processors like an intel 486 or one of the modern complex micrcontrollers. Plus rolling your own board layout is much easier as the timing is a lot more forgiving.

Disclosure: the 8051 was how I learned about computers back in the days when those things were pretty standard microcontrollers for many applications. And yes, right now I'm feeling old

[coppice]

That's wrong, there were actually quite a few scopes that could do single shot 1GSa/s the early '90s, for example the HP 54510A (my work horse back then).

I didn't realise that was single shot, I assumed it must've been equivalent time. That must've been quite something back then.

Well, such high sample rates weren't as common as they are today but it wasn't particularly exceptional (the 54500 Series was HP's mid-range scope series). There also was the older 54111D (1GSa/s with two channels or 2GSa/s on a single channel; came out in 1989).

Tektronix had some 1GSa/s real-time scopes at that time.

I think people tend to look at what they can buy off the shelf, and assume that is the current state of the art. What you can buy as a standard part is largely determined by there being a big enough market, willing to pay a high RE, to justify the NRE and marketing costs of launching the thing. Instrument makers are usually well ahead of people like Analog Devices and TI in the highest speed ADCs, because they are basically the only market for these things, and they can justify the entire life cycle costs of developing them. It took DSPs and FPGAs fast enough to crunch through the output of a 1Gsps converter before the incentive to make off the shelf ones was there.

[coppice]

Disclosure: the 8051 was how I learned about computers back in the days when those things were pretty standard microcontrollers for many applications. And yes, right now I'm feeling old

If the 8051 was around when you started, then you either started late in life, or you are not that old. :-)

[nctnico]

I'm not a fan of the 8051 because it has way too many weird things. Programming it in C also means keeping a lot of the 8051 limitations in mind during programming. Better use a real CPU with a normal data + address bus, real stack, etc.

[Wuerstchenhund]

I think people tend to look at what they can buy off the shelf, and assume that is the current state of the art. What you can buy as a standard part is largely determined by there being a big enough market, willing to pay a high RE, to justify the NRE and marketing costs of launching the thing. Instrument makers are usually well ahead of people like Analog Devices and TI in the highest speed ADCs, because they are basically the only market for these things, and they can justify the entire life cycle costs of developing them. It took DSPs and FPGAs fast enough to crunch through the output of a 1Gsps converter before the incentive to make off the shelf ones was there.

You're absolutely right, it's T&M manufacturers who drive ADC development for their scopes and signal analyzers, and who invest a lot of money in pushing them further. Mass market component manufacturers like AD or TI don't, they're only interested in large volumes, and therefore what they're offering is mostly standard tech and not state of the art.

I mean, what's the fastest 8bit ADC you can get on the component market these days, 4GSa/s or something like that? In scopes we're currently at 240GSa/s.

[Wuerstchenhund]

Disclosure: the 8051 was how I learned about computers back in the days when those things were pretty standard microcontrollers for many applications. And yes, right now I'm feeling old

If the 8051 was around when you started, then you either started late in life, or you are not that old. :-)

When I started the thing was actually called MCS51. Later we went to the 8085. So maybe I'm not that old after all (although much closer to retirement than to graduation) 

[Wuerstchenhund]

I'm not a fan of the 8051 because it has way too many weird things. Programming it in C also means keeping a lot of the 8051 limitations in mind during programming. Better use a real CPU with a normal data + address bus, real stack, etc.

Well, I wouldn't suggest to program an 8051 in a High Level Language like C, which just adds unnecessary complexity itself. And if the aim was to built a computer to run C programs then I'd agree. But don't forget that the OP wants to learn about the internal workings of computers, and a low complexity microcontroller programmed in Assembler is much easier than to use a complex CPU, add all the glue logic like memory controller to it, and then run interpreted or compiled programs on it.

[nctnico]

A Z80 isn't difficult to use. This is a Z80 board I made over 20 years ago:

At that time I only had a 20MHz ananalog scope and a multimeter to get it working. This was also the first PCB design I made using a CAD package; I know the component placement is far from optimal. I actually placed the Z80 between the RAM and the EPROM not realising the RAM and EPROM have almost identical connections so routing would have been much easier.

It only takes a decoder like the 74x138 and some OR gates (74x32) to decode the read/write strobes for the memory. The Z80 has a seperate I/O bus so an extra decoder allows to create more I/O space (which is where the extra chips and headers on the board are for).

Connecting an 8051 to external memory takes the same amount of glue logic and much more assembly programming to access that external memory! The Z80 really is the simplest choice here.

[Wuerstchenhund]

A Z80 isn't difficult to use.

Yes, the Z80 could be a good alternative (I thought you meant newer/more complex CPUs).

[Howardlong]

Another vote for the Z80 if you want to know about busses at what not.

Although it wasn't the first hand wired computer I built, my first Z80 was hand wired on stripboard in the 70s, and I got it working without a scope, just an LED plus resistor as a logic probe and a crap 1k ohm per volt multimeter were all I had as test equipment. Occasionally I might use a monostable and LED to show if something was oscillating or not if the pulse was so narrow it couldn't be discerned visually on an LED from always on or always off, but usually an LED plus resistor was all you needed.

True, I wouldn't want to do it that way nowadays, but TE was in relative terms really very expensive. Occasionally I'd have access to a scope, but it was rare. I remember the first scope I used with a delayed timebase, that really was something when trying to dubug your code at bus level, which is what we did back in those days without debuggers, logic analysers or in circuit emulators. When initially getting a machine up for the first time, we used to set a trigger reference point, and go through each of the control, data and address bus bits one at a time and mark down the binary on paper. You could then see the machine code and where things weren't happening as expected.

[tggzzz]

Another vote for the Z80 if you want to know about busses at what not.

Although it wasn't the first hand wired computer I built, my first Z80 was hand wired on stripboard in the 70s, and I got it working without a scope, just an LED plus resistor as a logic probe and a crap 1k ohm per volt multimeter were all I had as test equipment. Occasionally I might use a monostable and LED to show if something was oscillating or not if the pulse was so narrow it couldn't be discerned visually on an LED from always on or always off, but usually an LED plus resistor was all you needed.

True, I wouldn't want to do it that way nowadays, but TE was in relative terms really very expensive. Occasionally I'd have access to a scope, but it was rare. I remember the first scope I used with a delayed timebase, that really was something when trying to dubug your code at bus level, which is what we did back in those days without debuggers, logic analysers or in circuit emulators. When initially getting a machine up for the first time, we used to set a trigger reference point, and go through each of the control, data and address bus bits one at a time and mark down the binary on paper. You could then see the machine code and where things weren't happening as expected.

Just so, except I used a 6800 and avoided stripboard in favour of homebrew PCBs made with bizarre etch resists

[JohnPen]

I did similar with a Z80 in the 70s.  I did however have a home built copy of a 10 Mhz Telequipment Transistorized scope to use.  I made my own hand drawn PCBs for the Z80 system.  Main problem I had was forgetting I had inverted the enable signal on the separate Dynamic ram board and inverting it elsewhere as well to feed that board.  It worked some of the time in that mode but  was very unreliable and tended to forget everything in the RAM!  Programming was all in a 1K 2708 UVROM in machine code.  Later development moved on to Tiny Basic and Forth.  Certainly an easier chip to run with as it has it's own clock and single rail use. 

John

[Howardlong]

Indeed, back in those days it was common to need to supply multiple supply rails and multi-phase clocks, often at non-TTL levels, which complicated things. The Z80 largely simplified all that. However, getting the integrated dynamic RAM interface to work in spec with the DRAM of the day was another piece of work altogether. RC and gate propagation delays seemed to be the order of the day to get the timings just right.

[nbritton]

on paper the 2072A offers over double the spec:

at 3x the price - does not seen equal

If 2x the price I would say the odds are even

I just don't get it when someone compares a $400 scope with a $1200 one ??

The DS2072A is not $1,200, it is $789 over at tequipment.net with the eevblog discount, the price difference compared to the DS1054Z is $414.

DS1054Z: $375
MSO1074Z: $785
   MSO Option = $410

DS2072A: $789
MSO2072A: $1,165
   MSO Option = $376

From what I hear the MSO option on the DS1054Z is worthless, so you would need to buy a DS1054Z plus a separate logic analyzer. The Saleae Logic Pro 16 costs $599. Now you're up to $974. For $191 more you can upgrade to the MSO2072A.

[Fungus]

Sounds good. My recommendation then would be to forget about Raspberry Pis, ATMegas and other modern day complex stuff and start with something simple like an 8051.

Nah, I wouldn't user an 8051. It has built in memory, etc. If you're doing that then you might as well use something current like an AVR chip. Putting an AVR chip on a breadboard isn't much of a challenge though (plug it in, power it up, load a program - they just work!)

If I wanted to build a small computer with external buses I'd go for a Z80 or 6502. They need external RAM, ROM, I/O chips, etc.

Frequencies in the 1-4MHz range also don't need carefully designed PCBs to make them work. You can use perfboard and bits of wire.

[Howardlong]

on paper the 2072A offers over double the spec:

at 3x the price - does not seen equal

If 2x the price I would say the odds are even

I just don't get it when someone compares a $400 scope with a $1200 one ??

The DS2072A is not $1,200, it is $789 over at tequipment.net with the eevblog discount, the price difference compared to the DS1054Z is $414.

DS1054Z: $375
MSO1074Z: $785
   MSO Option = $410

DS2072A: $789
MSO2072A: $1,165
   MSO Option = $376

From what I hear the MSO option on the DS1054Z is worthless, so you would need to buy a DS1054Z plus a separate logic analyzer. The Saleae Logic Pro 16 costs $599. Now you're up to $974. For $191 more you can upgrade to the MSO2072A.

Because you were actually referring to the MSO2072A about which that comment was made (https://www.eevblog.com/forum/testgear/best-entry-level-oscilloscope-for-computer-engineering/msg753532/#msg753532):

When I woke up this morning I was heavy leaning towards an MSO2072A

Not sure what information you're referring to regarding the MSO on the DS1054Z: as far as I know, the DS1054Z doesn't have an MSO option. There was some talk recently that they might be introducing a version of the DS1000Z that might accept an after market MSO option, but I don't believe they're available as yet. At present, the only MSO option for the DS1000Z series on on the MSO1074Z and MSO1100Z.

The MSO on the MSO1074Z is OK in my experience.

What's not fantastic is the serial protocol decode (analogue or digital channels) because you can only decode and list what's on the screen, not what's in memory, and even then it undersamples the underlying display memory. In practice what this means is that although the scope might have an enormous buffer, you can't export a CSV of the decoded frames of the entire buffer, just the few bytes it can decode on the display. The protocol decode doesn't work on waveform recorded segments either (aka segmented memory). Although you can trigger on a given serial scenario (which does work with waveform record), you can't search on an already captured waveform. In I2C, it sometimes makes mistakes in decoding, although the I2C trigger works correctly. Having said that, it's without doubt better than no decode or trigger at all.

I've never used a DS/MSO2000A series so I'm afraid I can't comment on any limitations of protocol decoding on those.

[alank2]

From what I hear the MSO option on the DS1054Z is worthless, so you would need to buy a DS1054Z plus a separate logic analyzer. The Saleae Logic Pro 16 costs $599. Now you're up to $974. For $191 more you can upgrade to the MSO2072A.

That would be true only if a MSO2072A is _an upgrade_ from a DS1054Z + Saleae Logic Pro 16.  I don't think it is.  I am not against a MSO, but my experience is that you can't accomplish the same work on a MSO that you can with a USB LA like the Saleae or Intronix LogicPort.

[coppice]

I mean, what's the fastest 8bit ADC you can get on the component market these days, 4GSa/s or something like that? In scopes we're currently at 240GSa/s.

For general purpose chips the maximum sample rate of merchant parts is not that high. However, some specialist stuff is much faster. For example, you can find a presentation on the web about Fujitsu's 56Gsps and 80Gsps 8 bits ADCs for ethernet applications. Most of the chip is trying to deal with the never ending deluge of data flooding from the the converter. That's why it ends up as a specialist part, and not a pure ADC.

[nctnico]

From what I hear the MSO option on the DS1054Z is worthless, so you would need to buy a DS1054Z plus a separate logic analyzer. The Saleae Logic Pro 16 costs $599. Now you're up to $974. For $191 more you can upgrade to the MSO2072A.

That would be true only if a MSO2072A is _an upgrade_ from a DS1054Z + Saleae Logic Pro 16.  I don't think it is.  I am not against a MSO, but my experience is that you can't accomplish the same work on a MSO that you can with a USB LA like the Saleae or Intronix LogicPort.

Depends on the usage scenario but in general an MSO can do 95% of the logic analysis tasks. USB logic analysers are better at protocol decoding than low end MSOs like the ones from Rigol and Siglent. In case you need to do more complicated logic analysis tasks (complex triggers / higher speeds) getting a real logic analyser is a better choice because you'll have deeper memory, proper probes and better triggering capabilities. IMHO the Intronix Logicport is a bit of a toy due to the extremely short memory. You can make do ofcourse but I ran into it's limitations rather quickly.