Lampucera vs ?

[Chris1967]

Ναι αλλά με τα καλύτερα ρολόγια

https://www.aliexpress.com/item/1005005 ... 1802MBAvo2

[Dr Pan K]

Να κάνω ενα break down του κοστους στην παρούσα μορφή του.

Κουτί- σχάρα- φιλτρο ρεύματος- βυσματα πλάτης 170ευρω
Lampucera Dac kit περίπου 120 ευρω. Περιλαμβάνει την κεντρική πλακετα και το buffer καθως και μια τριπλή τροφοδοσία με αστείους μετασχηματιστες.

3 τοροειδεις περίπου 100 ευρω.

4 πλακετες τροφοδοσίας. Θα μπορουσαν να ειναι και 3. Καπου στα 90 ευρω με τα μεταφορικά

Αλλαγή τελεστικου με κορυφαιο 1612 περίπου 10 ευρω, οι λυχνίες απο την αλλη εχουν πάνω απο 300 αλλα έρχεται με κάποιες απλές κια δεν χρειάζεται την απόλυτη υπερβολή.

Ενα μικρο ποσο για καλωδια, αποστάτες κτλ. Το κουτι και η σχαρα θέλουν τρύπημα κτλ, αυτο προϋποθέτει ενα βασικό εργαστήριο.

Αν καποιος ενδιαφέρεται για τέτοια υλοποίηση μπορω να προτείνω συγκεκριμένα εξαρτήματα, λαμπες κουτια κτλ. Στο νημα υπάρχουν όλα αλλα λίγο απλωμένα.

[angpsi]

Άρα περίπου €500 χωρίς λάμπες, άντε €650 αν υπόλογίσει κανείς το converter του Chris για streaming. Μια και το πρότεινες, δεν βάζεις και τη λίστα συγκεντρωτικά με links για να το κλείσουμε για οποιονδήποτε ενδιαφερθεί; Προσπάθησα να το κάνω εγώ, αλλά νομίζω ότι κάτι χάνω. Αν παίζει όπως λέτε, και μόνο για τον χαβαλέ ίσως αξίζει κανείς να ασχοληθεί.

Btw, σε αυτά τα λεφτά, η σύγκριση με το Wavelight είναι άνευ αντικειμένου. Βεβαίως, αν η διαφορά στην ακουστική ποιότητα είναι σημαντικά μικρότερη από την τεράστια διαφορά του κόστους, τότε θα πρέπει ν' αρχίσουν να κλείνουν σπίτια στην κοινότητα του Hi-End.

[Dr Pan K]

Εδώ και καμια εβδομάδα συζητάμε με τον DimDim και τον Chris εαν αξίζει να ασχοληθούμε περαιτέρω.

Βασικα με τον Dim η κουβέντα ειναι η εξής, ποσο θα κοστιζε να φτιαξουμε ενα Dac board με το 4397, τελεστικο (1612) στη εξοδο (δλδ το ίδιο layout με την κινέζικη πλακετα) αλλα με ενσωματωμένο ενα XMOS 208 για receiver. Θα έμπαιναν επανω συμπαθητικά υλικα, γραμμικα τοπικά reg και θα γλιτωναμε απο την ανάγκη ενός εξωτερικου USB to Coaxial receiver/ reclocker buffer σαν αυτο που έφερε ο Chris.

Αυτο το dac θα κοστίσει καπου στα 200 ευρω (μαζι με 2 συμπαθητικά ρολογια, το τσιπακι του xmos, το τσίπ 4397, εκτυπωμενες πλακετες, reg, πυκνωτες κτλ).

Αν βάλουμε κουτί, σχαρα, φιλτρο εισόδου, 3 τοροειδεις με τις αντίστοιχες πλακετες τροφοδοσίας, την πλακετα του buffer με τις λυχνίες, βυσματα, ποδαρακια και καλωδιώσεις νομίζω σαν σύνολο θα παει στα 700 ευρω.

Και βέβαια ειναι επίσημη κατασκευή, πρεπει να συνδεθούν όλα αυτά αφού τρυπηθουν κουτια και σχάρες κτλ. Δεν ξερω πόσοι πραγματικά μπορούν και ταυτόχρονα επιθυμούν να φτιάξουν κατι τέτοιο.

[Στέλιος Περγιουδάκης]

Για την κεντρική πλακέτα υποθέτω εννοείς έτοιμη κολλημένη από εργοστάσιο σύμφωνα με τις οδηγίες του Δημήτρη, γιατί αυτή δεν κολλιέται από τον μέσο ντιουι.

[Dr Pan K]

Για την κεντρική πλακέτα υποθέτω εννοείς έτοιμη κολλημένη από εργοστάσιο σύμφωνα με τις οδηγίες του Δημήτρη, γιατί αυτή δεν κολλιέται από τον μέσο ντιουι.

Στέλιο το συζητήσαμε με τον Δημήτρη, αν τελικα προχωρησει (που δεν το ξερω) και βρεθουμε καμιά δεκαριά (που ισως και να είναι ευκολο) τότε η πλακετα θα έρθει populated, έτοιμη σε όλα και θα πρεπει να μπουν μονο τροφοδοσίες και κουτι. Που και αυτά δεν ειναι για όλους αλλα λέμε τωρα.

[Στέλιος Περγιουδάκης]

Η πλακέτα θα είναι και με το σχετικό λαμπάτο μπαφερ και ίσως θα χρειαστεί να κολληθούν μόνο τα υλικά τρύπας, φαντάζομαι.
Καθόλου καλώδια διασύνδεσης, μόνο απ τους μετ/τες.

[Dr Pan K]

Η πλακέτα θα είναι και με το σχετικό λαμπάτο μπαφερ και ίσως θα χρειαστεί να κολληθούν μόνο τα υλικά τρύπας, φαντάζομαι.
Καθόλου καλώδια διασύνδεσης, μόνο απ τους μετ/τες.

Στελιο σκεφτόμουν να μπει ξεχωριστή πλακετα για το buffer. Η πιο απλη λύση ειναι να αγοράσει ο καθενας οτι θελει και ισως κάποιοι να φτιάξουν δική τους. Σχέδια υπάρχουν πολλά και η δικιά μου ειναι απλη για να την κάνουμε reverse engineer.

[Dr Pan K]

Έκανα μια μικρη έρευνα όσον αφορά τα dac chip της Cirrus Logic. Βρήκα κάποια ενδιαφέροντα πράγματα. Να πω πως την lampucera την ειχα αγοράσει ςπο φιλο πριν καμια δεκαπενταριά χρόνια και παντα μου εκανε εντυπωση η φυσικότητα και ο πλούσιος ηχος της.

Τα διαφορα μοντέλα των dac της Cirrus Logic μπορούμε να τα χωρίσουμε σε δύο βασικές κατηγορίες, τα παλιά (43ΧΧ) και τα πιο νέα με πενταψηφιο κωδικό όπως το 43198. Το συγκεκριμένο ειναι η άμεση εξέλιξη του 4399 με ακομα καλυτερα τεχνικα χαρακτηριστικά (επιπέδου AKM τέταρτης γενιας και ESS Sabre)

Καπου εκει αρχίζει το ενδιαφερον. Τα παλιά τσιπ παιζουν καλυτερα παροτι μετράνε χειρότερα. Ο λόγος (τα απλοποιω οσο μπορω) ειναι ο τρόπος με τον οποίο διαχειρίζονται τον θορυβο, συγκεκριμένα ποσο μακριά εκτός του ακουστικού φάσματος καταλήγει το υψισυχνο σκουπίδι

Από Chat. Θα βάλω σε bold κάποια σημεία που εχουν ενδιαφέρον.

Yes. I found the measurements you're referring to, and there's an important distinction: the conspicuous feature is primarily the ultrasonic noise-shaping rise above ~40 kHz, not a set of large in-band harmonic distortion products.

CS43198 FFT
This is an actual measured FFT of a CS43198-based implementation, extending out to 200 kHz:

The characteristic shape is:

1 kHz fundamental at the left/center of the spectrum.

H2/H3 and higher harmonics are extremely low.

The noise floor is relatively flat through the audio band.

At roughly 40 kHz and above, the noise floor begins rising substantially.

It continues rising toward the upper part of the measurement bandwidth.

That observation is independently reported by several CS43198 measurements. One designer measured H2 ≈ −125 dB and H3 ≈ −128 dB at 1 kHz/1.7 Vrms, while another measured the evaluation board at better than −130 dB THD.

The particularly interesting measurement
A more revealing test is to put a 30 kHz full-scale tone alongside a −60 dBFS 1 kHz tone.

The measured result on the related CS43131 was:

1 kHz −60 dBFS alone

→ DR ≈ 129.8 dBA

1 kHz −60 dBFS + 30 kHz 0 dBFS

→ DR ≈ 110.3 dBA

The noise floor rises dramatically when the large ultrasonic signal is present.

That's the phenomenon I think you were remembering.

But I need to be careful here: those particular plots are CS43131, not CS43198. The two share closely related Cirrus architecture, but it would be wrong to present the CS43131 measurement as a CS43198 measurement.

What about the CS43198 specifically?
There is direct CS43198 evidence that the out-of-band noise is elevated above ~40 kHz. A designer who measured both the Cirrus evaluation board and his own CS43198 implementation specifically noted the higher OOB noise above 40 kHz and that selecting Wide Band Flatness Mode reduces it, but introduces tradeoffs in distortion.

And this is consistent with the CS43198's architecture: the chip uses aggressive noise shaping, so some quantization/noise energy is intentionally pushed outside the audio band.

Here's the key point about your original question
I wouldn't call this “distortion caused by oversampling.”

It's more accurately:

ΔΣ noise shaping → ultrasonic noise energy → interaction with the analog/output stage → potentially increased IMD or noise when strong ultrasonic content is present.

That's quite different from seeing, say:

1 kHz → H2 at 2 kHz → H3 at 3 kHz → H4 at 4 kHz

at unusually high levels.

In fact, the CS43198 can have exceptionally low H2/H3. A properly implemented unit has been measured around −125/−128 dB for those harmonics.

And this is where CS4399 becomes interesting
The CS4399 is also an oversampled multibit ΔΣ DAC, so it isn't immune to the fundamental mechanism. Cirrus explicitly describes its modulator as an oversampled multibit modulator with mismatch shaping.

What I haven't found is a credible apples-to-apples FFT showing that the CS43198 has significantly worse in-band distortion than the CS4399 under identical conditions.

So I would not yet conclude that the CS4399 is intrinsically cleaner.

The more defensible conclusion from the measurements is:

CS43198

spectacular in-band H2/H3 performance

very low THD under the specified conditions

pronounced ultrasonic noise-shaping behavior above ~40 kHz

filter/mode choice can affect the tradeoff

CS4399

also very low distortion

−108 dB specified THD+N versus −115 dB for CS43198

also uses noise-shaped ΔΣ architecture

I haven't found evidence showing the same CS43198-style problem is absent from it

If your concern is using one of these in a very high-end line-level DAC with an external analog stage, I'd actually focus on the 20 Hz–100 kHz FFT and IMD behavior, rather than the normal 20 Hz–20 kHz THD+N number. That's where the CS43198's architecture becomes interesting.


4397 vs 4399?

If you mean CS4397 vs CS4399, that's an interesting comparison because the 4397 is much older, but it has a couple of characteristics that may be relevant to what you're investigating.

CS4397 CS4399
Generation ~1999/2000 ~2018
Architecture Multibit ΔΣ + DEM 32-bit multibit ΔΣ + mismatch shaping
Resolution 24-bit 32-bit
Max PCM 192 kHz 384 kHz
Dynamic range 120 dB 130 dBA
THD+N −100 dB −108 dB
DSD Yes DSD64/128
Digital filters Internal + external interpolation option 5 selectable internal filters
Output Fully differential Differential
External reference Yes Internal
Supply ~5 V Low-voltage architecture
Integrated headphone amp No No
The CS4397's architecture is particularly interesting: Cirrus used a multibit ΔΣ modulator followed by dynamic-element matching (DEM) driving a multielement switched-capacitor DAC. The datasheet specifically claims that DEM gives low noise/distortion across signal levels and lower out-of-band noise than traditional 1-bit designs.

The CS4399 is a much more sophisticated modern implementation, with a 32-bit oversampled multibit modulator and mismatch-shaping technology. It achieves the considerably better published −108 dB THD+N / 130 dBA DR figures.

But here's the interesting part for your investigation
I would not automatically choose the 4399 over the 4397 if your priority is ultrasonic behavior.

The 4397 datasheet explicitly emphasizes: lower out-of-band noise and its architecture uses DEM + multielement switched-capacitor DAC.

That's potentially quite relevant to the issue we were discussing with the 43198.

The CS4399, meanwhile, is explicitly a newer oversampled multibit ΔΣ design, with five interpolation-filter choices and a much higher nominal performance envelope.

So I'd characterize them like this:

CS4397

Older technology

Lower headline resolution/performance

Very interesting architecture

Explicit emphasis on low OOB noise

External interpolation-filter capability

5-V analog environment

Could be an interesting candidate if you're trying to minimize ultrasonic ΔΣ artifacts

CS4399

Much better measured/spec'd conventional audio performance

32/384

Better THD+N

Better dynamic range

More sophisticated digital filtering

More compact/lower-power implementation

But doesn't necessarily mean it has the best ultrasonic behavior

And one correction to keep our discussion precise: neither is a true non-ΔΣ DAC. The CS4397 is explicitly described as an oversampled multibit delta-sigma modulator.

If you're chasing the “least nasty DAC architecture” rather than the best datasheet THD+N, I think the next comparison I'd make is CS4397 vs CS4398 vs CS4399 vs CS43198, specifically looking at FFT harmonic structure + ultrasonic noise + IMD, rather than their headline specs. That would be much more revealing.


Yesterday 3:54 PM
Do that comparison

I dug into this more carefully. The useful comparison is not the headline THD+N; it's architecture, out-of-band noise, and what the available FFT/measurement evidence actually shows.

One caveat up front: I could not find a trustworthy, controlled set of four identical FFT measurements (same analyzer, output stage, level, sample rate and filter) for CS4397/4398/4399/43198. So I won't manufacture a side-by-side FFT. We can, however, draw some fairly strong conclusions from the datasheets and published measurements.

1. Architecture
CS4397 CS4398 CS4399 CS43198
Era ~1999 ~2005 ~2018 ~2018
Modulator Multibit ΔΣ Multibit ΔΣ 32-bit multibit ΔΣ 32-bit multibit ΔΣ
Element matching DEM DEM / multibit architecture Mismatch shaping Mismatch shaping
PCM 24/192 24/192 32/384 32/384
Published DR 120 dB 120 dB 130 dBA 130 dBA
Published THD+N −100 dB ~−107 dB −108 dB −115 dB
DSD Yes Yes DSD128 DSD256
External interpolation Yes No No No
The CS4397 is unusual because Cirrus explicitly says its multibit architecture gives “significantly lower out-of-band noise” than traditional 1-bit designs, and it uses DEM to randomize/match the DAC elements.

The CS4399 and CS43198 move to Cirrus's newer mismatch-shaping implementation. Cirrus specifically says the latter eliminates distortion from on-chip component mismatch.

2. The 1-kHz FFT / harmonic picture
This is where the newer chips look spectacular.

For a properly implemented CS43198, measurements have reached roughly −130 dB THD at 1 kHz/0 dBFS on the evaluation board. However, other implementations have measured substantially worse, around −109 dB, demonstrating how sensitive the complete implementation can be.

The CS4399's specified result is:

−108 dB THD+N @ 0 dBFS, 20 Hz–20 kHz

and CS4398 is around:

−107 dB THD+N

while CS4397 is:

−100 dB THD+N.

So for ordinary in-band 1-kHz THD, the ranking is approximately:

CS43198 > CS4399 ≈ CS4398 > CS4397

But that's not the whole story.

3. Out-of-band noise is the really interesting part
Here's where your original suspicion about the 43198 becomes much more justified.

Published measurements of the CS43198 show an earlier rise in out-of-band noise, around/above 40 kHz. An engineer measuring the CS43198 evaluation board specifically reported higher noise above 40 kHz and noted that Wide Band Flatness Mode reduces the OOB noise but introduces other distortion tradeoffs.

Another measurement discussion describes the CS43198 as having an earlier rise of OOB noise than parts such as the AK4493, and notes that this can make it less attractive for measurement applications despite excellent audio-band performance.

So conceptually, the FFT looks something like:

dBFS
0 ┤ │
-20 ┤ │ 1 kHz fundamental
-40 ┤ │
-60 ┤ │
-80 ┤ │
-100┤ │
-120┤ │________________
-130┤ \
-140┤ \
0 20 40 60 80 100 kHz
↑
OOB noise rises
That rising ultrasonic floor is not harmonic distortion.

It's predominantly the result of the ΔΣ noise-shaping process pushing quantization noise out of the audio band.

4. Why CS4397 is particularly interesting
This is the part I find most relevant to what you're looking for.

The CS4397's datasheet explicitly emphasizes:

multibit architecture → lower out-of-band noise

and its block diagram is:

PCM → interpolation → multibit ΔΣ → DEM → multielement switched-capacitor DAC

rather than a simple 1-bit architecture.

It also allows an external interpolation filter.

That's unusual and potentially very valuable if your goal is controlling exactly what ultrasonic content reaches the analog stage.

Its published numbers are less impressive:

120 dB DR

−100 dB THD+N

but those numbers don't tell you how benign its ultrasonic spectrum is.

5. CS4398
The CS4398 is probably the sweet spot if you want an older Cirrus multibit design without going all the way back to the 4397.

Published figures:

120 dB A-weighted DR

117 dB unweighted DR

−107 dB THD+N

192 kHz PCM

differential output

Its datasheet specifies THD+N measurements over 10 Hz–20 kHz, so that number tells us very little about what happens above 20 kHz.

And this is important: CS4398 has remained popular in professional audio interfaces, which is a pretty good indication that its overall behavior is considered robust rather than merely impressive on a marketing spec sheet. Experienced measurement engineers have specifically described CS4398 as having stood the test of time.

6. CS4399
The CS4399 is where Cirrus moved substantially toward the modern low-power architecture.

Its headline numbers are excellent:

130 dBA DR
−108 dB THD+N
32/384

and it has five selectable interpolation-filter responses.

But note something subtle:

CS4399's −108 dB THD+N is only 1 dB better than CS4398's ~−107 dB.

Yet Cirrus advertises substantially more dynamic range and 32-bit/384-kHz capability.

So if we're asking:

Is CS4399 intrinsically dramatically better sounding than CS4398?

The specifications don't support that conclusion.

7. CS43198
This is the extreme version.

Cirrus took the CS4399 architecture and evolved it:

CS4399 → CS43198

with:

THD+N: −108 → −115 dB

DSD128 → DSD256

output increased to 2 Vrms

32/384 retained.

On paper it's the winner.

But the interesting measurement caveat is:

the CS43198 appears to push the noise floor upward sooner outside the audio band.

And that's exactly the point raised by engineers using it for instrumentation: phenomenal in-band performance, but OOB noise is the weak point.

My ranking for your particular criterion
If you're asking:

"Which Cirrus DAC has the least objectionable ΔΣ behavior?"
rather than:

"Which has the best THD+N specification?"
I'd rank them differently:

1. CS4398 — probably my choice

2. CS4397 — potentially excellent, especially if you exploit external filtering

3. CS4399 — technically excellent but less compelling if OOB behavior is the priority

4. CS43198 — best in-band numbers, but the most interesting OOB tradeoff

That's not a claim that the CS43198 sounds worse. It's a ranking for a very specific engineering objective: minimizing unwanted ultrasonic/noise-shaping consequences around the analog stage.

The really important experiment
If I were designing the DAC you're contemplating, I'd test these four with exactly the same analog stage:

A. 1 kHz @ −1 dBFS FFT, 0–100 kHz

B. 1 kHz @ −60 dBFS FFT, 0–100 kHz

C. 19 + 20 kHz IMD

D. 10 kHz + 30 kHz IMD

E. 1 kHz −60 dBFS + 20/30/40 kHz full-scale interferer

That last test is particularly revealing.

It answers the question:

Does the DAC's ultrasonic noise-shaping energy actually modulate the audio-band signal?

And that's much more meaningful than −115 vs −108 dB THD+N.

Bottom line
If you're building a high-end line DAC with an excellent external I/V/output stage, I would not automatically pick the CS43198 despite its spectacular −115 dB specification.

I'd seriously consider the CS4398, and I'd be very interested in the CS4397 if you can source good chips and are willing to design the interpolation/analog filtering carefully

[Dr Pan K]

Οποτε καπου 15 χρονια μετα έμαθα γιατι μου αρεσε παντα τόσο πολυ ο ηχος του παλιού 4397. Μπορεί να μην εχει τα απόλυτα τεχνικα χαρακτηριστικά στα χαρτιά αλλα εχει την καλύτερη συμπεριφορα σε θέματα διαχείρισης θορύβου.

Η εξέλιξη των specs ηρθε εις βαρος της μουσικης απόλαυσης. Τον πόλεμο τον κέρδισε η ESS, παρολαυτα ακομα δεν εχω ακούσει ενα dac της που να μου αρεσε. Θα προτιμήσω την αρχαία Lampucera μου.