• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Fosi Audio Box X5 Phono Preamp Review

Rate this phono stage:

  • 1. Poor (headless panther)

    Votes: 3 1.6%
  • 2. Not terrible (postman panther)

    Votes: 8 4.1%
  • 3. Fine (happy panther)

    Votes: 34 17.6%
  • 4. Great (golfing panther)

    Votes: 148 76.7%

  • Total voters
    193
Just to confirm I bought one from Amazon UK. No audible buzz from transformer (unless I put my ear directly on the PSU - very normal level). No audible hiss at lowest gain with my VM95ML and SL-100C and normal playback volumes.
 
This is the AP dashboard when a short plug/AP generator/dummycartridge is connected to the Box X5 input.
There we have it, again. 240uV (10Hz ...22kHz) noise with cartridge, but only 28uV with AP as source, about 18dB worse and SINAD would take a similar hit.
The points to consider is that the range from 1kHz to 20kHz is ~20x as large as the range from DC to 1kHz but the logarithmic frequency axis obscures this and makes the difference look less severe than it actually is.

Thanks for the measurement, perfectly confirming previous measurements from @SIY as well as calculations/simulations (EDIT: in my sim I had omitted the combined load capacitance but which in reality is there and often forms a peaking lowpass -- and the noise also undergoes than peaking and lowpass curve, that's why we see that significant bump at ~9kHz).
 
Last edited:
There we have it, again. 240uV (10Hz ...22kHz) noise with cartridge, but only 28uV with AP as source, about 18dB worse and SINAD would take a similar hit.
...
Finally strict on topic, see below. But here are some blank numbers from my LTSpice simulation at 20°C, as we are dealing with thermal noise.

Self noise of a quite standard MM 1kOhm in series with 500mH inductance, MM-pre input is 47kOhm in parallel to 350pF, resonance at 12kHz.

12k: 28nV/sqrtHz < that isn't nothing, the self noise without any amp the pickup just laying around sucking energy from the 'thermal bath' around
9k: 22nV/sqrtHz
6k: 14nV/sqrtHz
3k: 7,3nV/sqrtHz
1k: 4,5nV/sqrtHz

Impedance, to be used to calculate voltage noise as Noise current x Impedance => resulting (additional) noise voltage

12k: 45kOhm
9k: 35kOhm
6k: 22kOhm
3k: 10kOhm
1k: 3,1kOhm

Uncorrelated noise contributions sum by sqrt[ sqr(A) + sqr(B) + sqr(C) + ... ].

Example for different op amps:
NE5532 (Vn ~4nV/sqrtHz, In ~ 0,5pA/sqrtHz) at 9kHz => sqrt[ 22n^2 + (35k*0,5p)^2 + 4n^2 ]V/sqrtHZ = 28nV/sqrtHz
OPA1642 (Vn ~ 1nV/sqrtHz, In ~ 2pA/sqrtHz) at 9kHz => sqrt[ 22n^2 + (35k*2p)^2 + 1n^2 ]V/sqrtHZ = 73nV/sqrtHz

That gives, at that narrow band, some lumpy 8dB worse results for the OPA1642.

From that I concluded, that using an OPA1642 for MM was not the very best idea in the world. I changed, sigh, my verdict one step lower, you win. Otherwise, especially for the package as it is, the product is still outstanding. Be told, the difference will be washed away by the record's self noise by factors of magnitude once the needle hits the surface.
 
Last edited:
opa1612..not 1642...
;-)
You win again, price: you recalc all the numbers!!

Seriously, all that discussion is on, or even over the edge. Not only that the record, or even the mastering process would clearly underperform in comparison to the X5's low noise, still accounting for the 'current noise' contribution. It is, considering the analog fiesta once in a while at home far beyond any real use, the debate I mean.
 
You win again, price: you recalc all the numbers!!

Seriously, all that discussion is on, or even over the edge. Not only that the record, or even the mastering process would clearly underperform in comparison to the X5's low noise, still accounting for the 'current noise' contribution. It is, considering the analog fiesta once in a while at home far beyond any real use, the debate I mean.
not me who measures ... and classifies by "sinad at 1k in mm" these prephonos etc ;-)
(ps the question is to know to what extent degrades a given signal whether it comes from a phono, a tuner, a dac, a cd, a tape... the nature of signal etc is not really the subject... it seems to me... ;-))
 
Seriously, all that discussion is on, or even over the edge. Not only that the record, or even the mastering process would clearly underperform in comparison to the X5's low noise, still accounting for the 'current noise' contribution. It is, considering the analog fiesta once in a while at home far beyond any real use, the debate I mean.
This is true for almost any piece of gear, for almost all performance criteria. Basically, any review simply could have one boolean bottom line: "good enough" or not.

But ASR itself established a crazy race for best specs on the foundation of solid measurements... which is what went wrong here. While the shorted input SINAD gives useful data, it is not a good measure for the true actual perfomance and skews all those IMHO silly ranking charts. You can only go freaking wild with rankings when the underlying data is thorough and a fair competition is guaranteed.
 
This is true for almost any piece of gear, for almost all performance criteria. Basically, any review simply could have one boolean bottom line: "good enough" or not.
I like to see that "good enough".
But ASR itself established a crazy race ...
... based on solid measurements, right. My point was, that using an MM generator, simulated, as input for the test wasn't correct either. Because of the basic thermal self-noise of such a thing. With the somehow ideal NE5532 as an input opamp for MM the self noise of the MM dominates easily.

The question arises, what to run after in the test ;-) I leave it to you.

A more ideal architecture for X(y) would have looked like:

MC: dedicated cinch input, 100Ohm or 1000Ohm || 1nF or 10nF, linear opa1612 => ..
MM: dedicated cinch input, 47kOhm || 50pF or 150pF, linear NE5532A (socketed, changeable) => ...

... => common passive RIAA => linear opa1612 with variable gain => output
 
maybe just dominated by a "marketing" aspect...
the aop 1612 is now selling in the general public "top modern opa", (much highlighted by this site elsewhere)
and
the idea of being able to play with the aop "5532 on dip support" too... !

""two selling ideas""....
;-)
 
Last edited:
Does your generator do an inverse RIAA sweep? Looking at your overload sweeps, reference the 1kHz sensitivity, the signal coming out the cart at 20 Hz is 20 dB lower. So if you take it as 5mV at 1kHz, at 20Hz it will be 500uV while at 20 kHz it will be 50mV. This seems to indicate the Fosi has 26 dB o/load at 20 Hz (very good) and 15 dB at 20 kHz. But, why are your 10kHz and 20kHz traces overlaid? I would expect there to be a 6 dB difference.
This is the frequency response, but is it what you wanted?
I don't speak English and I use browser translation, so I'm not sure if I understand it correctly.
 

Attachments

  • 80_fosi_fr.png
    80_fosi_fr.png
    99.9 KB · Views: 53
  • 81_eq_inverse_riaa.png
    81_eq_inverse_riaa.png
    95 KB · Views: 49
... based on solid measurements, right. My point was, that using an MM generator, simulated, as input for the test wasn't correct either. Because of the basic thermal self-noise of such a thing. With the somehow ideal NE5532 as an input opamp for MM the self noise of the MM dominates easily.
Not sure what to make of this.

Of course there is thermal "self" noise from the resistive part of MM cartridge impedance but since that resistance is low, usually below 1kOhm, thus below 4nV/rtHz. But the low resistance does not help at higher frequencies where the inductance isolates this resistance, allowing more of the thermal noise of the 47k load resistance (which is part of the amp, not the source) to come through. The load resistor can be replaced by an "electrically cold resistor", that is, a much higher resistor like 470k (10x) with the "ground" terminal driven to -9x the input voltage, this reduces self noise considerably. All in all this represent the input noise baseline which leads to an output noise baseline (for a totally noise-less amp) with the RIAA gain in place.

What we are interested in is the amount of noise additionally introduced by a real amp on top of that noise baseline. If it has only voltage noise (like a good FET-input opamp, preferably with a low 1/f corner**), then the noise penalty from the amp will be absolutely minimal as long as it is below a couple of nV/rtHz. Once we have noise current from the amp input, the noise voltage delevoped accross the total source impedance adds to the thermal noise (in an uncorrelated fashion) and this can very quickly be the dominant contributor not only to the added noise but also to total noise.

**) four paralleled sections of an OPA164x are pretty much ideal, 2.5nV/rtHz, much better than the ~4nV/rtHz of a 5532 plus ~1pA/rtHz noise current (as per TI datasheet). And even a single section (5nV/trtHz) will gibe better results than 5532, let alone OPA1612.

So, in the shorted-input SINAD, reduced effective input noise from a cold resistor (aka synthesized input impedance) is not reflected.
Much more importantly, current-noise induced additional noise is not reflected as well.
 
Of course there is thermal "self" noise from the resistive part of MM cartridge impedance but since that resistance is low, ...
I personally think, and famous designer D.Self gave me the hint, and the LTSpice simulation confirms, that the thermal noise is bigger. It originates, as it seems, in the 47kOhm load, while the inductor isolates the 1kOhm of the MM itself. I see the 47kOhm as part of the generator ;-) And all in all it is a resonant circuit.

With 28nV/sqrtHz tops the contribution of the MM (and its necessary surroundings) itself is not nil. Keep it added to the measurment when using a simulated source, or substract it once known?! What then does the measurement on noise really represent?

Yes, we could parallel even JFET opamps. Diminishing returns, and again, what are we after?

Again strict on topic: The use of the opamps in the X5 is not ideal, admitted, but not a deal breaker either. When using RIAA equalization, lowering the treble by 40dB (!!), the numbers given above change again in favor of keeping the OPA1612 in place, not worrying about the current noise, actually. The criticism, as I feel it, is a bit academic, if you can say that.
 
Again strict on topic: The use of the opamps in the X5 is not ideal, admitted, but not a deal breaker either. When using RIAA equalization, lowering the treble by 40dB (!!), the numbers given above change again in favor of keeping the OPA1612 in place, not worrying about the current noise, actually. The criticism, as I feel it, is a bit academic, if you can say that.
that's not the subject in the background...
there is just a problem in certain cases with this protocol which makes it not very relevant in mm..
this little fosi is just an opportunity to point it out by very competent people...
and to just think of an approach on a more representative load "and ap" for mm for the future ....

leads to relativizing the classifications "sinad 1k at db "of mm measurements practiced here...*

it's great to have such competent people here...
thank you to them (for their educational and technical efforts..)

*We must not be naive about the impact of these measures on many people's purchasing decisions.for some it really is the "tables of the law"...( """ until the next....""") ;-)
 
Last edited:
that's not the subject in the background...
there is just a problem in certain cases with this protocol which makes it not very relevant in mm..
As you are very keen on numbers, opamp types ;-)
We are talking about numbers, that represent something real. The numbers quantify a mind model of the thing in focus. The model came out to be incomplete. The noise current is not considered.

Now as the current shall be converted to voltage at the MM impedance, which is fair use, we have to acknowledge, that the impedance by itself, due to thermal activity of the comprising electrons, adds an own noise contribution.

Eventually we land at a standard (tbd) MM, sum all noise contributions up and get a maximum S/N ratio for phono that cannot be superseeded.

The record surface is considered noiseless, or better to say, the S/N ratio as tested is only valid, if the device is not in use. Full circle ;-)

Edit 0: we could introduce a 'noise gate' that shuts off the phono pre output once it doesn't detect the usual clicks and pops anymore
Edit 1: some turntables shorten the output to ground once the tonearm is lifted

Problem solved, halfway serious.
 
Last edited:
Maybe some expert could explain to me, why the headroom estimations of @nagster show opposite results. Increasing headroom with frequency, as I would expect if the second (or third?) amplification stage was the limiting factor.

Why is the input capacitance changing in nagster‘s measurements, when switching to MC? As far as I see there is only one physical C present.

(Not as if it was relevant, but running MC with 38dB of amplification is not foreseen?)
My take, from experience, is that this is likely a measurement artefact. With 100R//150pF at 1kHz, the influence of the capacitor is extremely low, the resistor dominates the impedance by 80dB (10000x) at 1kHz. And given the low test voltage of 10mV, even the best LCR meter will have issues.
With test frequency = 100kHz and 100mV level I'm pretty sure the capactance will close in on the value we see in MM setting. There, you can see the effect of test frequency and test level as well. The 20kHz/100mV measurement is the closest one to reality. With 47k//150pF at 20kHz, the cap and resistor contribute about equal parts to the impedance and the meter can easily calculate resisitve and capacitive parts with high precision (7 significant digits is still completely ridiculuous, though).
Agree. The anomaly at the lowest test frequency is a major clue (think overload).

C meters are not a particularly reliable way to do this measurement- to get accurate and repeatable numbers, you need subterfuge to only tickle the preamp with a low level signal. IME, the best way to go is to inject a square wave via a voltage divider, measure the rise time at the input (using a 10x probe!), and using the Thevenin resistance, calculate C from rise time divided by Thevenin resistance.
I re-measured the input capacity of the Box X5. The clip lead only goes up to 100kHz so I changed it just to be safe.
I'm a bit concerned about the 44kohm at 56DB MM 100kHz, but is it generally as expected?
 

Attachments

  • 100_fosi_lcr_.jpg
    100_fosi_lcr_.jpg
    206.8 KB · Views: 82
  • 101_fosi_lcr_38DB.png
    101_fosi_lcr_38DB.png
    39.4 KB · Views: 92
  • 102_fosi_lcr_56DB.png
    102_fosi_lcr_56DB.png
    45.2 KB · Views: 88
Hi i want to buy one unit to plug it to my bluesound icon and benefit to dirac correction for my records. Is it good with a 2m blue ortofon. I've read the thread but kinda got lost in all measurements aha
 
Hi i want to buy one unit to plug it to my bluesound icon and benefit to dirac correction for my records. Is it good with a 2m blue ortofon. I've read the thread but kinda got lost in all measurements aha
beside all the measurements, as far as I read here everybody who has the unit is satisfied. your risk is just 100 bucks...
 
Back
Top Bottom